Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * handling kvm guest interrupts
4 *
5 * Copyright IBM Corp. 2008, 2020
6 *
7 * Author(s): Carsten Otte <cotte@de.ibm.com>
8 */
9
10#define KMSG_COMPONENT "kvm-s390"
11#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
12
13#include <linux/interrupt.h>
14#include <linux/kvm_host.h>
15#include <linux/hrtimer.h>
16#include <linux/mmu_context.h>
17#include <linux/nospec.h>
18#include <linux/signal.h>
19#include <linux/slab.h>
20#include <linux/bitmap.h>
21#include <linux/vmalloc.h>
22#include <asm/asm-offsets.h>
23#include <asm/dis.h>
24#include <linux/uaccess.h>
25#include <asm/sclp.h>
26#include <asm/isc.h>
27#include <asm/gmap.h>
28#include <asm/switch_to.h>
29#include <asm/nmi.h>
30#include <asm/airq.h>
31#include <asm/tpi.h>
32#include "kvm-s390.h"
33#include "gaccess.h"
34#include "trace-s390.h"
35#include "pci.h"
36
37#define PFAULT_INIT 0x0600
38#define PFAULT_DONE 0x0680
39#define VIRTIO_PARAM 0x0d00
40
41static struct kvm_s390_gib *gib;
42
43/* handle external calls via sigp interpretation facility */
44static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
45{
46 int c, scn;
47
48 if (!kvm_s390_test_cpuflags(vcpu, CPUSTAT_ECALL_PEND))
49 return 0;
50
51 BUG_ON(!kvm_s390_use_sca_entries());
52 read_lock(&vcpu->kvm->arch.sca_lock);
53 if (vcpu->kvm->arch.use_esca) {
54 struct esca_block *sca = vcpu->kvm->arch.sca;
55 union esca_sigp_ctrl sigp_ctrl =
56 sca->cpu[vcpu->vcpu_id].sigp_ctrl;
57
58 c = sigp_ctrl.c;
59 scn = sigp_ctrl.scn;
60 } else {
61 struct bsca_block *sca = vcpu->kvm->arch.sca;
62 union bsca_sigp_ctrl sigp_ctrl =
63 sca->cpu[vcpu->vcpu_id].sigp_ctrl;
64
65 c = sigp_ctrl.c;
66 scn = sigp_ctrl.scn;
67 }
68 read_unlock(&vcpu->kvm->arch.sca_lock);
69
70 if (src_id)
71 *src_id = scn;
72
73 return c;
74}
75
76static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id)
77{
78 int expect, rc;
79
80 BUG_ON(!kvm_s390_use_sca_entries());
81 read_lock(&vcpu->kvm->arch.sca_lock);
82 if (vcpu->kvm->arch.use_esca) {
83 struct esca_block *sca = vcpu->kvm->arch.sca;
84 union esca_sigp_ctrl *sigp_ctrl =
85 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
86 union esca_sigp_ctrl new_val = {0}, old_val;
87
88 old_val = READ_ONCE(*sigp_ctrl);
89 new_val.scn = src_id;
90 new_val.c = 1;
91 old_val.c = 0;
92
93 expect = old_val.value;
94 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
95 } else {
96 struct bsca_block *sca = vcpu->kvm->arch.sca;
97 union bsca_sigp_ctrl *sigp_ctrl =
98 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
99 union bsca_sigp_ctrl new_val = {0}, old_val;
100
101 old_val = READ_ONCE(*sigp_ctrl);
102 new_val.scn = src_id;
103 new_val.c = 1;
104 old_val.c = 0;
105
106 expect = old_val.value;
107 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
108 }
109 read_unlock(&vcpu->kvm->arch.sca_lock);
110
111 if (rc != expect) {
112 /* another external call is pending */
113 return -EBUSY;
114 }
115 kvm_s390_set_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
116 return 0;
117}
118
119static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
120{
121 int rc, expect;
122
123 if (!kvm_s390_use_sca_entries())
124 return;
125 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
126 read_lock(&vcpu->kvm->arch.sca_lock);
127 if (vcpu->kvm->arch.use_esca) {
128 struct esca_block *sca = vcpu->kvm->arch.sca;
129 union esca_sigp_ctrl *sigp_ctrl =
130 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
131 union esca_sigp_ctrl old;
132
133 old = READ_ONCE(*sigp_ctrl);
134 expect = old.value;
135 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
136 } else {
137 struct bsca_block *sca = vcpu->kvm->arch.sca;
138 union bsca_sigp_ctrl *sigp_ctrl =
139 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
140 union bsca_sigp_ctrl old;
141
142 old = READ_ONCE(*sigp_ctrl);
143 expect = old.value;
144 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
145 }
146 read_unlock(&vcpu->kvm->arch.sca_lock);
147 WARN_ON(rc != expect); /* cannot clear? */
148}
149
150int psw_extint_disabled(struct kvm_vcpu *vcpu)
151{
152 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
153}
154
155static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
156{
157 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
158}
159
160static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
161{
162 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
163}
164
165static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
166{
167 return psw_extint_disabled(vcpu) &&
168 psw_ioint_disabled(vcpu) &&
169 psw_mchk_disabled(vcpu);
170}
171
172static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
173{
174 if (psw_extint_disabled(vcpu) ||
175 !(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
176 return 0;
177 if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
178 /* No timer interrupts when single stepping */
179 return 0;
180 return 1;
181}
182
183static int ckc_irq_pending(struct kvm_vcpu *vcpu)
184{
185 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
186 const u64 ckc = vcpu->arch.sie_block->ckc;
187
188 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
189 if ((s64)ckc >= (s64)now)
190 return 0;
191 } else if (ckc >= now) {
192 return 0;
193 }
194 return ckc_interrupts_enabled(vcpu);
195}
196
197static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
198{
199 return !psw_extint_disabled(vcpu) &&
200 (vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK);
201}
202
203static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
204{
205 if (!cpu_timer_interrupts_enabled(vcpu))
206 return 0;
207 return kvm_s390_get_cpu_timer(vcpu) >> 63;
208}
209
210static uint64_t isc_to_isc_bits(int isc)
211{
212 return (0x80 >> isc) << 24;
213}
214
215static inline u32 isc_to_int_word(u8 isc)
216{
217 return ((u32)isc << 27) | 0x80000000;
218}
219
220static inline u8 int_word_to_isc(u32 int_word)
221{
222 return (int_word & 0x38000000) >> 27;
223}
224
225/*
226 * To use atomic bitmap functions, we have to provide a bitmap address
227 * that is u64 aligned. However, the ipm might be u32 aligned.
228 * Therefore, we logically start the bitmap at the very beginning of the
229 * struct and fixup the bit number.
230 */
231#define IPM_BIT_OFFSET (offsetof(struct kvm_s390_gisa, ipm) * BITS_PER_BYTE)
232
233/**
234 * gisa_set_iam - change the GISA interruption alert mask
235 *
236 * @gisa: gisa to operate on
237 * @iam: new IAM value to use
238 *
239 * Change the IAM atomically with the next alert address and the IPM
240 * of the GISA if the GISA is not part of the GIB alert list. All three
241 * fields are located in the first long word of the GISA.
242 *
243 * Returns: 0 on success
244 * -EBUSY in case the gisa is part of the alert list
245 */
246static inline int gisa_set_iam(struct kvm_s390_gisa *gisa, u8 iam)
247{
248 u64 word, _word;
249
250 do {
251 word = READ_ONCE(gisa->u64.word[0]);
252 if ((u64)gisa != word >> 32)
253 return -EBUSY;
254 _word = (word & ~0xffUL) | iam;
255 } while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
256
257 return 0;
258}
259
260/**
261 * gisa_clear_ipm - clear the GISA interruption pending mask
262 *
263 * @gisa: gisa to operate on
264 *
265 * Clear the IPM atomically with the next alert address and the IAM
266 * of the GISA unconditionally. All three fields are located in the
267 * first long word of the GISA.
268 */
269static inline void gisa_clear_ipm(struct kvm_s390_gisa *gisa)
270{
271 u64 word, _word;
272
273 do {
274 word = READ_ONCE(gisa->u64.word[0]);
275 _word = word & ~(0xffUL << 24);
276 } while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
277}
278
279/**
280 * gisa_get_ipm_or_restore_iam - return IPM or restore GISA IAM
281 *
282 * @gi: gisa interrupt struct to work on
283 *
284 * Atomically restores the interruption alert mask if none of the
285 * relevant ISCs are pending and return the IPM.
286 *
287 * Returns: the relevant pending ISCs
288 */
289static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi)
290{
291 u8 pending_mask, alert_mask;
292 u64 word, _word;
293
294 do {
295 word = READ_ONCE(gi->origin->u64.word[0]);
296 alert_mask = READ_ONCE(gi->alert.mask);
297 pending_mask = (u8)(word >> 24) & alert_mask;
298 if (pending_mask)
299 return pending_mask;
300 _word = (word & ~0xffUL) | alert_mask;
301 } while (cmpxchg(&gi->origin->u64.word[0], word, _word) != word);
302
303 return 0;
304}
305
306static inline int gisa_in_alert_list(struct kvm_s390_gisa *gisa)
307{
308 return READ_ONCE(gisa->next_alert) != (u32)(u64)gisa;
309}
310
311static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
312{
313 set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
314}
315
316static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa)
317{
318 return READ_ONCE(gisa->ipm);
319}
320
321static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
322{
323 return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
324}
325
326static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu)
327{
328 unsigned long pending = vcpu->kvm->arch.float_int.pending_irqs |
329 vcpu->arch.local_int.pending_irqs;
330
331 pending &= ~vcpu->kvm->arch.float_int.masked_irqs;
332 return pending;
333}
334
335static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
336{
337 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
338 unsigned long pending_mask;
339
340 pending_mask = pending_irqs_no_gisa(vcpu);
341 if (gi->origin)
342 pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7;
343 return pending_mask;
344}
345
346static inline int isc_to_irq_type(unsigned long isc)
347{
348 return IRQ_PEND_IO_ISC_0 - isc;
349}
350
351static inline int irq_type_to_isc(unsigned long irq_type)
352{
353 return IRQ_PEND_IO_ISC_0 - irq_type;
354}
355
356static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
357 unsigned long active_mask)
358{
359 int i;
360
361 for (i = 0; i <= MAX_ISC; i++)
362 if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
363 active_mask &= ~(1UL << (isc_to_irq_type(i)));
364
365 return active_mask;
366}
367
368static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
369{
370 unsigned long active_mask;
371
372 active_mask = pending_irqs(vcpu);
373 if (!active_mask)
374 return 0;
375
376 if (psw_extint_disabled(vcpu))
377 active_mask &= ~IRQ_PEND_EXT_MASK;
378 if (psw_ioint_disabled(vcpu))
379 active_mask &= ~IRQ_PEND_IO_MASK;
380 else
381 active_mask = disable_iscs(vcpu, active_mask);
382 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
383 __clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
384 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK))
385 __clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
386 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
387 __clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
388 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK))
389 __clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
390 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK)) {
391 __clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
392 __clear_bit(IRQ_PEND_EXT_SERVICE_EV, &active_mask);
393 }
394 if (psw_mchk_disabled(vcpu))
395 active_mask &= ~IRQ_PEND_MCHK_MASK;
396 /* PV guest cpus can have a single interruption injected at a time. */
397 if (kvm_s390_pv_cpu_get_handle(vcpu) &&
398 vcpu->arch.sie_block->iictl != IICTL_CODE_NONE)
399 active_mask &= ~(IRQ_PEND_EXT_II_MASK |
400 IRQ_PEND_IO_MASK |
401 IRQ_PEND_MCHK_MASK);
402 /*
403 * Check both floating and local interrupt's cr14 because
404 * bit IRQ_PEND_MCHK_REP could be set in both cases.
405 */
406 if (!(vcpu->arch.sie_block->gcr[14] &
407 (vcpu->kvm->arch.float_int.mchk.cr14 |
408 vcpu->arch.local_int.irq.mchk.cr14)))
409 __clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
410
411 /*
412 * STOP irqs will never be actively delivered. They are triggered via
413 * intercept requests and cleared when the stop intercept is performed.
414 */
415 __clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
416
417 return active_mask;
418}
419
420static void __set_cpu_idle(struct kvm_vcpu *vcpu)
421{
422 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
423 set_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
424}
425
426static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
427{
428 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
429 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
430}
431
432static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
433{
434 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT |
435 CPUSTAT_STOP_INT);
436 vcpu->arch.sie_block->lctl = 0x0000;
437 vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
438
439 if (guestdbg_enabled(vcpu)) {
440 vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
441 LCTL_CR10 | LCTL_CR11);
442 vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
443 }
444}
445
446static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
447{
448 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK))
449 return;
450 if (psw_ioint_disabled(vcpu))
451 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT);
452 else
453 vcpu->arch.sie_block->lctl |= LCTL_CR6;
454}
455
456static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
457{
458 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK))
459 return;
460 if (psw_extint_disabled(vcpu))
461 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
462 else
463 vcpu->arch.sie_block->lctl |= LCTL_CR0;
464}
465
466static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
467{
468 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK))
469 return;
470 if (psw_mchk_disabled(vcpu))
471 vcpu->arch.sie_block->ictl |= ICTL_LPSW;
472 else
473 vcpu->arch.sie_block->lctl |= LCTL_CR14;
474}
475
476static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
477{
478 if (kvm_s390_is_stop_irq_pending(vcpu))
479 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
480}
481
482/* Set interception request for non-deliverable interrupts */
483static void set_intercept_indicators(struct kvm_vcpu *vcpu)
484{
485 set_intercept_indicators_io(vcpu);
486 set_intercept_indicators_ext(vcpu);
487 set_intercept_indicators_mchk(vcpu);
488 set_intercept_indicators_stop(vcpu);
489}
490
491static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
492{
493 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
494 int rc = 0;
495
496 vcpu->stat.deliver_cputm++;
497 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
498 0, 0);
499 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
500 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
501 vcpu->arch.sie_block->eic = EXT_IRQ_CPU_TIMER;
502 } else {
503 rc = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
504 (u16 *)__LC_EXT_INT_CODE);
505 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
506 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
507 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
508 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
509 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
510 }
511 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
512 return rc ? -EFAULT : 0;
513}
514
515static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
516{
517 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
518 int rc = 0;
519
520 vcpu->stat.deliver_ckc++;
521 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
522 0, 0);
523 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
524 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
525 vcpu->arch.sie_block->eic = EXT_IRQ_CLK_COMP;
526 } else {
527 rc = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
528 (u16 __user *)__LC_EXT_INT_CODE);
529 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
530 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
531 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
532 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
533 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
534 }
535 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
536 return rc ? -EFAULT : 0;
537}
538
539static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
540{
541 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
542 struct kvm_s390_ext_info ext;
543 int rc;
544
545 spin_lock(&li->lock);
546 ext = li->irq.ext;
547 clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
548 li->irq.ext.ext_params2 = 0;
549 spin_unlock(&li->lock);
550
551 VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
552 ext.ext_params2);
553 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
554 KVM_S390_INT_PFAULT_INIT,
555 0, ext.ext_params2);
556
557 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
558 rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
559 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
560 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
561 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
562 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
563 rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
564 return rc ? -EFAULT : 0;
565}
566
567static int __write_machine_check(struct kvm_vcpu *vcpu,
568 struct kvm_s390_mchk_info *mchk)
569{
570 unsigned long ext_sa_addr;
571 unsigned long lc;
572 freg_t fprs[NUM_FPRS];
573 union mci mci;
574 int rc;
575
576 /*
577 * All other possible payload for a machine check (e.g. the register
578 * contents in the save area) will be handled by the ultravisor, as
579 * the hypervisor does not not have the needed information for
580 * protected guests.
581 */
582 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
583 vcpu->arch.sie_block->iictl = IICTL_CODE_MCHK;
584 vcpu->arch.sie_block->mcic = mchk->mcic;
585 vcpu->arch.sie_block->faddr = mchk->failing_storage_address;
586 vcpu->arch.sie_block->edc = mchk->ext_damage_code;
587 return 0;
588 }
589
590 mci.val = mchk->mcic;
591 /* take care of lazy register loading */
592 save_fpu_regs();
593 save_access_regs(vcpu->run->s.regs.acrs);
594 if (MACHINE_HAS_GS && vcpu->arch.gs_enabled)
595 save_gs_cb(current->thread.gs_cb);
596
597 /* Extended save area */
598 rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
599 sizeof(unsigned long));
600 /* Only bits 0 through 63-LC are used for address formation */
601 lc = ext_sa_addr & MCESA_LC_MASK;
602 if (test_kvm_facility(vcpu->kvm, 133)) {
603 switch (lc) {
604 case 0:
605 case 10:
606 ext_sa_addr &= ~0x3ffUL;
607 break;
608 case 11:
609 ext_sa_addr &= ~0x7ffUL;
610 break;
611 case 12:
612 ext_sa_addr &= ~0xfffUL;
613 break;
614 default:
615 ext_sa_addr = 0;
616 break;
617 }
618 } else {
619 ext_sa_addr &= ~0x3ffUL;
620 }
621
622 if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
623 if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
624 512))
625 mci.vr = 0;
626 } else {
627 mci.vr = 0;
628 }
629 if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
630 && (lc == 11 || lc == 12)) {
631 if (write_guest_abs(vcpu, ext_sa_addr + 1024,
632 &vcpu->run->s.regs.gscb, 32))
633 mci.gs = 0;
634 } else {
635 mci.gs = 0;
636 }
637
638 /* General interruption information */
639 rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
640 rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
641 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
642 rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
643 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
644 rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
645
646 /* Register-save areas */
647 if (MACHINE_HAS_VX) {
648 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
649 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
650 } else {
651 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
652 vcpu->run->s.regs.fprs, 128);
653 }
654 rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
655 vcpu->run->s.regs.gprs, 128);
656 rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
657 (u32 __user *) __LC_FP_CREG_SAVE_AREA);
658 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
659 (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
660 rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
661 (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
662 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
663 (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
664 rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
665 &vcpu->run->s.regs.acrs, 64);
666 rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
667 &vcpu->arch.sie_block->gcr, 128);
668
669 /* Extended interruption information */
670 rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
671 (u32 __user *) __LC_EXT_DAMAGE_CODE);
672 rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
673 (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
674 rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
675 sizeof(mchk->fixed_logout));
676 return rc ? -EFAULT : 0;
677}
678
679static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
680{
681 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
682 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
683 struct kvm_s390_mchk_info mchk = {};
684 int deliver = 0;
685 int rc = 0;
686
687 spin_lock(&fi->lock);
688 spin_lock(&li->lock);
689 if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
690 test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
691 /*
692 * If there was an exigent machine check pending, then any
693 * repressible machine checks that might have been pending
694 * are indicated along with it, so always clear bits for
695 * repressible and exigent interrupts
696 */
697 mchk = li->irq.mchk;
698 clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
699 clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
700 memset(&li->irq.mchk, 0, sizeof(mchk));
701 deliver = 1;
702 }
703 /*
704 * We indicate floating repressible conditions along with
705 * other pending conditions. Channel Report Pending and Channel
706 * Subsystem damage are the only two and are indicated by
707 * bits in mcic and masked in cr14.
708 */
709 if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
710 mchk.mcic |= fi->mchk.mcic;
711 mchk.cr14 |= fi->mchk.cr14;
712 memset(&fi->mchk, 0, sizeof(mchk));
713 deliver = 1;
714 }
715 spin_unlock(&li->lock);
716 spin_unlock(&fi->lock);
717
718 if (deliver) {
719 VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
720 mchk.mcic);
721 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
722 KVM_S390_MCHK,
723 mchk.cr14, mchk.mcic);
724 vcpu->stat.deliver_machine_check++;
725 rc = __write_machine_check(vcpu, &mchk);
726 }
727 return rc;
728}
729
730static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
731{
732 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
733 int rc = 0;
734
735 VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
736 vcpu->stat.deliver_restart_signal++;
737 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
738
739 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
740 vcpu->arch.sie_block->iictl = IICTL_CODE_RESTART;
741 } else {
742 rc = write_guest_lc(vcpu,
743 offsetof(struct lowcore, restart_old_psw),
744 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
745 rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
746 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
747 }
748 clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
749 return rc ? -EFAULT : 0;
750}
751
752static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
753{
754 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
755 struct kvm_s390_prefix_info prefix;
756
757 spin_lock(&li->lock);
758 prefix = li->irq.prefix;
759 li->irq.prefix.address = 0;
760 clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
761 spin_unlock(&li->lock);
762
763 vcpu->stat.deliver_prefix_signal++;
764 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
765 KVM_S390_SIGP_SET_PREFIX,
766 prefix.address, 0);
767
768 kvm_s390_set_prefix(vcpu, prefix.address);
769 return 0;
770}
771
772static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
773{
774 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
775 int rc;
776 int cpu_addr;
777
778 spin_lock(&li->lock);
779 cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
780 clear_bit(cpu_addr, li->sigp_emerg_pending);
781 if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
782 clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
783 spin_unlock(&li->lock);
784
785 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
786 vcpu->stat.deliver_emergency_signal++;
787 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
788 cpu_addr, 0);
789 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
790 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
791 vcpu->arch.sie_block->eic = EXT_IRQ_EMERGENCY_SIG;
792 vcpu->arch.sie_block->extcpuaddr = cpu_addr;
793 return 0;
794 }
795
796 rc = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
797 (u16 *)__LC_EXT_INT_CODE);
798 rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
799 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
800 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
801 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
802 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
803 return rc ? -EFAULT : 0;
804}
805
806static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
807{
808 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
809 struct kvm_s390_extcall_info extcall;
810 int rc;
811
812 spin_lock(&li->lock);
813 extcall = li->irq.extcall;
814 li->irq.extcall.code = 0;
815 clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
816 spin_unlock(&li->lock);
817
818 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
819 vcpu->stat.deliver_external_call++;
820 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
821 KVM_S390_INT_EXTERNAL_CALL,
822 extcall.code, 0);
823 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
824 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
825 vcpu->arch.sie_block->eic = EXT_IRQ_EXTERNAL_CALL;
826 vcpu->arch.sie_block->extcpuaddr = extcall.code;
827 return 0;
828 }
829
830 rc = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
831 (u16 *)__LC_EXT_INT_CODE);
832 rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
833 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
834 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
835 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
836 sizeof(psw_t));
837 return rc ? -EFAULT : 0;
838}
839
840static int __deliver_prog_pv(struct kvm_vcpu *vcpu, u16 code)
841{
842 switch (code) {
843 case PGM_SPECIFICATION:
844 vcpu->arch.sie_block->iictl = IICTL_CODE_SPECIFICATION;
845 break;
846 case PGM_OPERAND:
847 vcpu->arch.sie_block->iictl = IICTL_CODE_OPERAND;
848 break;
849 default:
850 return -EINVAL;
851 }
852 return 0;
853}
854
855static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
856{
857 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
858 struct kvm_s390_pgm_info pgm_info;
859 int rc = 0, nullifying = false;
860 u16 ilen;
861
862 spin_lock(&li->lock);
863 pgm_info = li->irq.pgm;
864 clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
865 memset(&li->irq.pgm, 0, sizeof(pgm_info));
866 spin_unlock(&li->lock);
867
868 ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
869 VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
870 pgm_info.code, ilen);
871 vcpu->stat.deliver_program++;
872 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
873 pgm_info.code, 0);
874
875 /* PER is handled by the ultravisor */
876 if (kvm_s390_pv_cpu_is_protected(vcpu))
877 return __deliver_prog_pv(vcpu, pgm_info.code & ~PGM_PER);
878
879 switch (pgm_info.code & ~PGM_PER) {
880 case PGM_AFX_TRANSLATION:
881 case PGM_ASX_TRANSLATION:
882 case PGM_EX_TRANSLATION:
883 case PGM_LFX_TRANSLATION:
884 case PGM_LSTE_SEQUENCE:
885 case PGM_LSX_TRANSLATION:
886 case PGM_LX_TRANSLATION:
887 case PGM_PRIMARY_AUTHORITY:
888 case PGM_SECONDARY_AUTHORITY:
889 nullifying = true;
890 fallthrough;
891 case PGM_SPACE_SWITCH:
892 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
893 (u64 *)__LC_TRANS_EXC_CODE);
894 break;
895 case PGM_ALEN_TRANSLATION:
896 case PGM_ALE_SEQUENCE:
897 case PGM_ASTE_INSTANCE:
898 case PGM_ASTE_SEQUENCE:
899 case PGM_ASTE_VALIDITY:
900 case PGM_EXTENDED_AUTHORITY:
901 rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
902 (u8 *)__LC_EXC_ACCESS_ID);
903 nullifying = true;
904 break;
905 case PGM_ASCE_TYPE:
906 case PGM_PAGE_TRANSLATION:
907 case PGM_REGION_FIRST_TRANS:
908 case PGM_REGION_SECOND_TRANS:
909 case PGM_REGION_THIRD_TRANS:
910 case PGM_SEGMENT_TRANSLATION:
911 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
912 (u64 *)__LC_TRANS_EXC_CODE);
913 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
914 (u8 *)__LC_EXC_ACCESS_ID);
915 rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
916 (u8 *)__LC_OP_ACCESS_ID);
917 nullifying = true;
918 break;
919 case PGM_MONITOR:
920 rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
921 (u16 *)__LC_MON_CLASS_NR);
922 rc |= put_guest_lc(vcpu, pgm_info.mon_code,
923 (u64 *)__LC_MON_CODE);
924 break;
925 case PGM_VECTOR_PROCESSING:
926 case PGM_DATA:
927 rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
928 (u32 *)__LC_DATA_EXC_CODE);
929 break;
930 case PGM_PROTECTION:
931 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
932 (u64 *)__LC_TRANS_EXC_CODE);
933 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
934 (u8 *)__LC_EXC_ACCESS_ID);
935 break;
936 case PGM_STACK_FULL:
937 case PGM_STACK_EMPTY:
938 case PGM_STACK_SPECIFICATION:
939 case PGM_STACK_TYPE:
940 case PGM_STACK_OPERATION:
941 case PGM_TRACE_TABEL:
942 case PGM_CRYPTO_OPERATION:
943 nullifying = true;
944 break;
945 }
946
947 if (pgm_info.code & PGM_PER) {
948 rc |= put_guest_lc(vcpu, pgm_info.per_code,
949 (u8 *) __LC_PER_CODE);
950 rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
951 (u8 *)__LC_PER_ATMID);
952 rc |= put_guest_lc(vcpu, pgm_info.per_address,
953 (u64 *) __LC_PER_ADDRESS);
954 rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
955 (u8 *) __LC_PER_ACCESS_ID);
956 }
957
958 if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
959 kvm_s390_rewind_psw(vcpu, ilen);
960
961 /* bit 1+2 of the target are the ilc, so we can directly use ilen */
962 rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
963 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
964 (u64 *) __LC_PGM_LAST_BREAK);
965 rc |= put_guest_lc(vcpu, pgm_info.code,
966 (u16 *)__LC_PGM_INT_CODE);
967 rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
968 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
969 rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
970 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
971 return rc ? -EFAULT : 0;
972}
973
974#define SCCB_MASK 0xFFFFFFF8
975#define SCCB_EVENT_PENDING 0x3
976
977static int write_sclp(struct kvm_vcpu *vcpu, u32 parm)
978{
979 int rc;
980
981 if (kvm_s390_pv_cpu_get_handle(vcpu)) {
982 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
983 vcpu->arch.sie_block->eic = EXT_IRQ_SERVICE_SIG;
984 vcpu->arch.sie_block->eiparams = parm;
985 return 0;
986 }
987
988 rc = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
989 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
990 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
991 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
992 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
993 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
994 rc |= put_guest_lc(vcpu, parm,
995 (u32 *)__LC_EXT_PARAMS);
996
997 return rc ? -EFAULT : 0;
998}
999
1000static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
1001{
1002 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1003 struct kvm_s390_ext_info ext;
1004
1005 spin_lock(&fi->lock);
1006 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs) ||
1007 !(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
1008 spin_unlock(&fi->lock);
1009 return 0;
1010 }
1011 ext = fi->srv_signal;
1012 memset(&fi->srv_signal, 0, sizeof(ext));
1013 clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1014 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1015 if (kvm_s390_pv_cpu_is_protected(vcpu))
1016 set_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs);
1017 spin_unlock(&fi->lock);
1018
1019 VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
1020 ext.ext_params);
1021 vcpu->stat.deliver_service_signal++;
1022 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1023 ext.ext_params, 0);
1024
1025 return write_sclp(vcpu, ext.ext_params);
1026}
1027
1028static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu)
1029{
1030 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1031 struct kvm_s390_ext_info ext;
1032
1033 spin_lock(&fi->lock);
1034 if (!(test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs))) {
1035 spin_unlock(&fi->lock);
1036 return 0;
1037 }
1038 ext = fi->srv_signal;
1039 /* only clear the event bit */
1040 fi->srv_signal.ext_params &= ~SCCB_EVENT_PENDING;
1041 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1042 spin_unlock(&fi->lock);
1043
1044 VCPU_EVENT(vcpu, 4, "%s", "deliver: sclp parameter event");
1045 vcpu->stat.deliver_service_signal++;
1046 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1047 ext.ext_params, 0);
1048
1049 return write_sclp(vcpu, SCCB_EVENT_PENDING);
1050}
1051
1052static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
1053{
1054 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1055 struct kvm_s390_interrupt_info *inti;
1056 int rc = 0;
1057
1058 spin_lock(&fi->lock);
1059 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
1060 struct kvm_s390_interrupt_info,
1061 list);
1062 if (inti) {
1063 list_del(&inti->list);
1064 fi->counters[FIRQ_CNTR_PFAULT] -= 1;
1065 }
1066 if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
1067 clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1068 spin_unlock(&fi->lock);
1069
1070 if (inti) {
1071 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1072 KVM_S390_INT_PFAULT_DONE, 0,
1073 inti->ext.ext_params2);
1074 VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
1075 inti->ext.ext_params2);
1076
1077 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1078 (u16 *)__LC_EXT_INT_CODE);
1079 rc |= put_guest_lc(vcpu, PFAULT_DONE,
1080 (u16 *)__LC_EXT_CPU_ADDR);
1081 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1082 &vcpu->arch.sie_block->gpsw,
1083 sizeof(psw_t));
1084 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1085 &vcpu->arch.sie_block->gpsw,
1086 sizeof(psw_t));
1087 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1088 (u64 *)__LC_EXT_PARAMS2);
1089 kfree(inti);
1090 }
1091 return rc ? -EFAULT : 0;
1092}
1093
1094static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
1095{
1096 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1097 struct kvm_s390_interrupt_info *inti;
1098 int rc = 0;
1099
1100 spin_lock(&fi->lock);
1101 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
1102 struct kvm_s390_interrupt_info,
1103 list);
1104 if (inti) {
1105 VCPU_EVENT(vcpu, 4,
1106 "deliver: virtio parm: 0x%x,parm64: 0x%llx",
1107 inti->ext.ext_params, inti->ext.ext_params2);
1108 vcpu->stat.deliver_virtio++;
1109 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1110 inti->type,
1111 inti->ext.ext_params,
1112 inti->ext.ext_params2);
1113 list_del(&inti->list);
1114 fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
1115 }
1116 if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
1117 clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1118 spin_unlock(&fi->lock);
1119
1120 if (inti) {
1121 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1122 (u16 *)__LC_EXT_INT_CODE);
1123 rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
1124 (u16 *)__LC_EXT_CPU_ADDR);
1125 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1126 &vcpu->arch.sie_block->gpsw,
1127 sizeof(psw_t));
1128 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1129 &vcpu->arch.sie_block->gpsw,
1130 sizeof(psw_t));
1131 rc |= put_guest_lc(vcpu, inti->ext.ext_params,
1132 (u32 *)__LC_EXT_PARAMS);
1133 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1134 (u64 *)__LC_EXT_PARAMS2);
1135 kfree(inti);
1136 }
1137 return rc ? -EFAULT : 0;
1138}
1139
1140static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io)
1141{
1142 int rc;
1143
1144 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
1145 vcpu->arch.sie_block->iictl = IICTL_CODE_IO;
1146 vcpu->arch.sie_block->subchannel_id = io->subchannel_id;
1147 vcpu->arch.sie_block->subchannel_nr = io->subchannel_nr;
1148 vcpu->arch.sie_block->io_int_parm = io->io_int_parm;
1149 vcpu->arch.sie_block->io_int_word = io->io_int_word;
1150 return 0;
1151 }
1152
1153 rc = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID);
1154 rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR);
1155 rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM);
1156 rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD);
1157 rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
1158 &vcpu->arch.sie_block->gpsw,
1159 sizeof(psw_t));
1160 rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
1161 &vcpu->arch.sie_block->gpsw,
1162 sizeof(psw_t));
1163 return rc ? -EFAULT : 0;
1164}
1165
1166static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
1167 unsigned long irq_type)
1168{
1169 struct list_head *isc_list;
1170 struct kvm_s390_float_interrupt *fi;
1171 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1172 struct kvm_s390_interrupt_info *inti = NULL;
1173 struct kvm_s390_io_info io;
1174 u32 isc;
1175 int rc = 0;
1176
1177 fi = &vcpu->kvm->arch.float_int;
1178
1179 spin_lock(&fi->lock);
1180 isc = irq_type_to_isc(irq_type);
1181 isc_list = &fi->lists[isc];
1182 inti = list_first_entry_or_null(isc_list,
1183 struct kvm_s390_interrupt_info,
1184 list);
1185 if (inti) {
1186 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1187 VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
1188 else
1189 VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
1190 inti->io.subchannel_id >> 8,
1191 inti->io.subchannel_id >> 1 & 0x3,
1192 inti->io.subchannel_nr);
1193
1194 vcpu->stat.deliver_io++;
1195 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1196 inti->type,
1197 ((__u32)inti->io.subchannel_id << 16) |
1198 inti->io.subchannel_nr,
1199 ((__u64)inti->io.io_int_parm << 32) |
1200 inti->io.io_int_word);
1201 list_del(&inti->list);
1202 fi->counters[FIRQ_CNTR_IO] -= 1;
1203 }
1204 if (list_empty(isc_list))
1205 clear_bit(irq_type, &fi->pending_irqs);
1206 spin_unlock(&fi->lock);
1207
1208 if (inti) {
1209 rc = __do_deliver_io(vcpu, &(inti->io));
1210 kfree(inti);
1211 goto out;
1212 }
1213
1214 if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) {
1215 /*
1216 * in case an adapter interrupt was not delivered
1217 * in SIE context KVM will handle the delivery
1218 */
1219 VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc);
1220 memset(&io, 0, sizeof(io));
1221 io.io_int_word = isc_to_int_word(isc);
1222 vcpu->stat.deliver_io++;
1223 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1224 KVM_S390_INT_IO(1, 0, 0, 0),
1225 ((__u32)io.subchannel_id << 16) |
1226 io.subchannel_nr,
1227 ((__u64)io.io_int_parm << 32) |
1228 io.io_int_word);
1229 rc = __do_deliver_io(vcpu, &io);
1230 }
1231out:
1232 return rc;
1233}
1234
1235/* Check whether an external call is pending (deliverable or not) */
1236int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
1237{
1238 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1239
1240 if (!sclp.has_sigpif)
1241 return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
1242
1243 return sca_ext_call_pending(vcpu, NULL);
1244}
1245
1246int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
1247{
1248 if (deliverable_irqs(vcpu))
1249 return 1;
1250
1251 if (kvm_cpu_has_pending_timer(vcpu))
1252 return 1;
1253
1254 /* external call pending and deliverable */
1255 if (kvm_s390_ext_call_pending(vcpu) &&
1256 !psw_extint_disabled(vcpu) &&
1257 (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
1258 return 1;
1259
1260 if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1261 return 1;
1262 return 0;
1263}
1264
1265int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1266{
1267 return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1268}
1269
1270static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1271{
1272 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1273 const u64 ckc = vcpu->arch.sie_block->ckc;
1274 u64 cputm, sltime = 0;
1275
1276 if (ckc_interrupts_enabled(vcpu)) {
1277 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
1278 if ((s64)now < (s64)ckc)
1279 sltime = tod_to_ns((s64)ckc - (s64)now);
1280 } else if (now < ckc) {
1281 sltime = tod_to_ns(ckc - now);
1282 }
1283 /* already expired */
1284 if (!sltime)
1285 return 0;
1286 if (cpu_timer_interrupts_enabled(vcpu)) {
1287 cputm = kvm_s390_get_cpu_timer(vcpu);
1288 /* already expired? */
1289 if (cputm >> 63)
1290 return 0;
1291 return min_t(u64, sltime, tod_to_ns(cputm));
1292 }
1293 } else if (cpu_timer_interrupts_enabled(vcpu)) {
1294 sltime = kvm_s390_get_cpu_timer(vcpu);
1295 /* already expired? */
1296 if (sltime >> 63)
1297 return 0;
1298 }
1299 return sltime;
1300}
1301
1302int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1303{
1304 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1305 u64 sltime;
1306
1307 vcpu->stat.exit_wait_state++;
1308
1309 /* fast path */
1310 if (kvm_arch_vcpu_runnable(vcpu))
1311 return 0;
1312
1313 if (psw_interrupts_disabled(vcpu)) {
1314 VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1315 return -EOPNOTSUPP; /* disabled wait */
1316 }
1317
1318 if (gi->origin &&
1319 (gisa_get_ipm_or_restore_iam(gi) &
1320 vcpu->arch.sie_block->gcr[6] >> 24))
1321 return 0;
1322
1323 if (!ckc_interrupts_enabled(vcpu) &&
1324 !cpu_timer_interrupts_enabled(vcpu)) {
1325 VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1326 __set_cpu_idle(vcpu);
1327 goto no_timer;
1328 }
1329
1330 sltime = __calculate_sltime(vcpu);
1331 if (!sltime)
1332 return 0;
1333
1334 __set_cpu_idle(vcpu);
1335 hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1336 VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1337no_timer:
1338 kvm_vcpu_srcu_read_unlock(vcpu);
1339 kvm_vcpu_halt(vcpu);
1340 vcpu->valid_wakeup = false;
1341 __unset_cpu_idle(vcpu);
1342 kvm_vcpu_srcu_read_lock(vcpu);
1343
1344 hrtimer_cancel(&vcpu->arch.ckc_timer);
1345 return 0;
1346}
1347
1348void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1349{
1350 vcpu->valid_wakeup = true;
1351 kvm_vcpu_wake_up(vcpu);
1352
1353 /*
1354 * The VCPU might not be sleeping but rather executing VSIE. Let's
1355 * kick it, so it leaves the SIE to process the request.
1356 */
1357 kvm_s390_vsie_kick(vcpu);
1358}
1359
1360enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1361{
1362 struct kvm_vcpu *vcpu;
1363 u64 sltime;
1364
1365 vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1366 sltime = __calculate_sltime(vcpu);
1367
1368 /*
1369 * If the monotonic clock runs faster than the tod clock we might be
1370 * woken up too early and have to go back to sleep to avoid deadlocks.
1371 */
1372 if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1373 return HRTIMER_RESTART;
1374 kvm_s390_vcpu_wakeup(vcpu);
1375 return HRTIMER_NORESTART;
1376}
1377
1378void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1379{
1380 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1381
1382 spin_lock(&li->lock);
1383 li->pending_irqs = 0;
1384 bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1385 memset(&li->irq, 0, sizeof(li->irq));
1386 spin_unlock(&li->lock);
1387
1388 sca_clear_ext_call(vcpu);
1389}
1390
1391int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1392{
1393 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1394 int rc = 0;
1395 unsigned long irq_type;
1396 unsigned long irqs;
1397
1398 __reset_intercept_indicators(vcpu);
1399
1400 /* pending ckc conditions might have been invalidated */
1401 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1402 if (ckc_irq_pending(vcpu))
1403 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1404
1405 /* pending cpu timer conditions might have been invalidated */
1406 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1407 if (cpu_timer_irq_pending(vcpu))
1408 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1409
1410 while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1411 /* bits are in the reverse order of interrupt priority */
1412 irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1413 switch (irq_type) {
1414 case IRQ_PEND_IO_ISC_0:
1415 case IRQ_PEND_IO_ISC_1:
1416 case IRQ_PEND_IO_ISC_2:
1417 case IRQ_PEND_IO_ISC_3:
1418 case IRQ_PEND_IO_ISC_4:
1419 case IRQ_PEND_IO_ISC_5:
1420 case IRQ_PEND_IO_ISC_6:
1421 case IRQ_PEND_IO_ISC_7:
1422 rc = __deliver_io(vcpu, irq_type);
1423 break;
1424 case IRQ_PEND_MCHK_EX:
1425 case IRQ_PEND_MCHK_REP:
1426 rc = __deliver_machine_check(vcpu);
1427 break;
1428 case IRQ_PEND_PROG:
1429 rc = __deliver_prog(vcpu);
1430 break;
1431 case IRQ_PEND_EXT_EMERGENCY:
1432 rc = __deliver_emergency_signal(vcpu);
1433 break;
1434 case IRQ_PEND_EXT_EXTERNAL:
1435 rc = __deliver_external_call(vcpu);
1436 break;
1437 case IRQ_PEND_EXT_CLOCK_COMP:
1438 rc = __deliver_ckc(vcpu);
1439 break;
1440 case IRQ_PEND_EXT_CPU_TIMER:
1441 rc = __deliver_cpu_timer(vcpu);
1442 break;
1443 case IRQ_PEND_RESTART:
1444 rc = __deliver_restart(vcpu);
1445 break;
1446 case IRQ_PEND_SET_PREFIX:
1447 rc = __deliver_set_prefix(vcpu);
1448 break;
1449 case IRQ_PEND_PFAULT_INIT:
1450 rc = __deliver_pfault_init(vcpu);
1451 break;
1452 case IRQ_PEND_EXT_SERVICE:
1453 rc = __deliver_service(vcpu);
1454 break;
1455 case IRQ_PEND_EXT_SERVICE_EV:
1456 rc = __deliver_service_ev(vcpu);
1457 break;
1458 case IRQ_PEND_PFAULT_DONE:
1459 rc = __deliver_pfault_done(vcpu);
1460 break;
1461 case IRQ_PEND_VIRTIO:
1462 rc = __deliver_virtio(vcpu);
1463 break;
1464 default:
1465 WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1466 clear_bit(irq_type, &li->pending_irqs);
1467 }
1468 }
1469
1470 set_intercept_indicators(vcpu);
1471
1472 return rc;
1473}
1474
1475static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1476{
1477 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1478
1479 vcpu->stat.inject_program++;
1480 VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1481 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1482 irq->u.pgm.code, 0);
1483
1484 if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1485 /* auto detection if no valid ILC was given */
1486 irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1487 irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1488 irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1489 }
1490
1491 if (irq->u.pgm.code == PGM_PER) {
1492 li->irq.pgm.code |= PGM_PER;
1493 li->irq.pgm.flags = irq->u.pgm.flags;
1494 /* only modify PER related information */
1495 li->irq.pgm.per_address = irq->u.pgm.per_address;
1496 li->irq.pgm.per_code = irq->u.pgm.per_code;
1497 li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1498 li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1499 } else if (!(irq->u.pgm.code & PGM_PER)) {
1500 li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1501 irq->u.pgm.code;
1502 li->irq.pgm.flags = irq->u.pgm.flags;
1503 /* only modify non-PER information */
1504 li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1505 li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1506 li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1507 li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1508 li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1509 li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1510 } else {
1511 li->irq.pgm = irq->u.pgm;
1512 }
1513 set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1514 return 0;
1515}
1516
1517static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1518{
1519 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1520
1521 vcpu->stat.inject_pfault_init++;
1522 VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1523 irq->u.ext.ext_params2);
1524 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1525 irq->u.ext.ext_params,
1526 irq->u.ext.ext_params2);
1527
1528 li->irq.ext = irq->u.ext;
1529 set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1530 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1531 return 0;
1532}
1533
1534static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1535{
1536 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1537 struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1538 uint16_t src_id = irq->u.extcall.code;
1539
1540 vcpu->stat.inject_external_call++;
1541 VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1542 src_id);
1543 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1544 src_id, 0);
1545
1546 /* sending vcpu invalid */
1547 if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1548 return -EINVAL;
1549
1550 if (sclp.has_sigpif && !kvm_s390_pv_cpu_get_handle(vcpu))
1551 return sca_inject_ext_call(vcpu, src_id);
1552
1553 if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1554 return -EBUSY;
1555 *extcall = irq->u.extcall;
1556 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1557 return 0;
1558}
1559
1560static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1561{
1562 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1563 struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1564
1565 vcpu->stat.inject_set_prefix++;
1566 VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1567 irq->u.prefix.address);
1568 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1569 irq->u.prefix.address, 0);
1570
1571 if (!is_vcpu_stopped(vcpu))
1572 return -EBUSY;
1573
1574 *prefix = irq->u.prefix;
1575 set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1576 return 0;
1577}
1578
1579#define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1580static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1581{
1582 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1583 struct kvm_s390_stop_info *stop = &li->irq.stop;
1584 int rc = 0;
1585
1586 vcpu->stat.inject_stop_signal++;
1587 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1588
1589 if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1590 return -EINVAL;
1591
1592 if (is_vcpu_stopped(vcpu)) {
1593 if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1594 rc = kvm_s390_store_status_unloaded(vcpu,
1595 KVM_S390_STORE_STATUS_NOADDR);
1596 return rc;
1597 }
1598
1599 if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1600 return -EBUSY;
1601 stop->flags = irq->u.stop.flags;
1602 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
1603 return 0;
1604}
1605
1606static int __inject_sigp_restart(struct kvm_vcpu *vcpu)
1607{
1608 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1609
1610 vcpu->stat.inject_restart++;
1611 VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1612 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1613
1614 set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1615 return 0;
1616}
1617
1618static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1619 struct kvm_s390_irq *irq)
1620{
1621 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1622
1623 vcpu->stat.inject_emergency_signal++;
1624 VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1625 irq->u.emerg.code);
1626 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1627 irq->u.emerg.code, 0);
1628
1629 /* sending vcpu invalid */
1630 if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1631 return -EINVAL;
1632
1633 set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1634 set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1635 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1636 return 0;
1637}
1638
1639static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1640{
1641 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1642 struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1643
1644 vcpu->stat.inject_mchk++;
1645 VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1646 irq->u.mchk.mcic);
1647 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1648 irq->u.mchk.mcic);
1649
1650 /*
1651 * Because repressible machine checks can be indicated along with
1652 * exigent machine checks (PoP, Chapter 11, Interruption action)
1653 * we need to combine cr14, mcic and external damage code.
1654 * Failing storage address and the logout area should not be or'ed
1655 * together, we just indicate the last occurrence of the corresponding
1656 * machine check
1657 */
1658 mchk->cr14 |= irq->u.mchk.cr14;
1659 mchk->mcic |= irq->u.mchk.mcic;
1660 mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1661 mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1662 memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1663 sizeof(mchk->fixed_logout));
1664 if (mchk->mcic & MCHK_EX_MASK)
1665 set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1666 else if (mchk->mcic & MCHK_REP_MASK)
1667 set_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
1668 return 0;
1669}
1670
1671static int __inject_ckc(struct kvm_vcpu *vcpu)
1672{
1673 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1674
1675 vcpu->stat.inject_ckc++;
1676 VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1677 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1678 0, 0);
1679
1680 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1681 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1682 return 0;
1683}
1684
1685static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1686{
1687 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1688
1689 vcpu->stat.inject_cputm++;
1690 VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1691 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1692 0, 0);
1693
1694 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1695 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1696 return 0;
1697}
1698
1699static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1700 int isc, u32 schid)
1701{
1702 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1703 struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1704 struct kvm_s390_interrupt_info *iter;
1705 u16 id = (schid & 0xffff0000U) >> 16;
1706 u16 nr = schid & 0x0000ffffU;
1707
1708 spin_lock(&fi->lock);
1709 list_for_each_entry(iter, isc_list, list) {
1710 if (schid && (id != iter->io.subchannel_id ||
1711 nr != iter->io.subchannel_nr))
1712 continue;
1713 /* found an appropriate entry */
1714 list_del_init(&iter->list);
1715 fi->counters[FIRQ_CNTR_IO] -= 1;
1716 if (list_empty(isc_list))
1717 clear_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1718 spin_unlock(&fi->lock);
1719 return iter;
1720 }
1721 spin_unlock(&fi->lock);
1722 return NULL;
1723}
1724
1725static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm,
1726 u64 isc_mask, u32 schid)
1727{
1728 struct kvm_s390_interrupt_info *inti = NULL;
1729 int isc;
1730
1731 for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1732 if (isc_mask & isc_to_isc_bits(isc))
1733 inti = get_io_int(kvm, isc, schid);
1734 }
1735 return inti;
1736}
1737
1738static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid)
1739{
1740 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1741 unsigned long active_mask;
1742 int isc;
1743
1744 if (schid)
1745 goto out;
1746 if (!gi->origin)
1747 goto out;
1748
1749 active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32;
1750 while (active_mask) {
1751 isc = __fls(active_mask) ^ (BITS_PER_LONG - 1);
1752 if (gisa_tac_ipm_gisc(gi->origin, isc))
1753 return isc;
1754 clear_bit_inv(isc, &active_mask);
1755 }
1756out:
1757 return -EINVAL;
1758}
1759
1760/*
1761 * Dequeue and return an I/O interrupt matching any of the interruption
1762 * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1763 * Take into account the interrupts pending in the interrupt list and in GISA.
1764 *
1765 * Note that for a guest that does not enable I/O interrupts
1766 * but relies on TPI, a flood of classic interrupts may starve
1767 * out adapter interrupts on the same isc. Linux does not do
1768 * that, and it is possible to work around the issue by configuring
1769 * different iscs for classic and adapter interrupts in the guest,
1770 * but we may want to revisit this in the future.
1771 */
1772struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1773 u64 isc_mask, u32 schid)
1774{
1775 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1776 struct kvm_s390_interrupt_info *inti, *tmp_inti;
1777 int isc;
1778
1779 inti = get_top_io_int(kvm, isc_mask, schid);
1780
1781 isc = get_top_gisa_isc(kvm, isc_mask, schid);
1782 if (isc < 0)
1783 /* no AI in GISA */
1784 goto out;
1785
1786 if (!inti)
1787 /* AI in GISA but no classical IO int */
1788 goto gisa_out;
1789
1790 /* both types of interrupts present */
1791 if (int_word_to_isc(inti->io.io_int_word) <= isc) {
1792 /* classical IO int with higher priority */
1793 gisa_set_ipm_gisc(gi->origin, isc);
1794 goto out;
1795 }
1796gisa_out:
1797 tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
1798 if (tmp_inti) {
1799 tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0);
1800 tmp_inti->io.io_int_word = isc_to_int_word(isc);
1801 if (inti)
1802 kvm_s390_reinject_io_int(kvm, inti);
1803 inti = tmp_inti;
1804 } else
1805 gisa_set_ipm_gisc(gi->origin, isc);
1806out:
1807 return inti;
1808}
1809
1810static int __inject_service(struct kvm *kvm,
1811 struct kvm_s390_interrupt_info *inti)
1812{
1813 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1814
1815 kvm->stat.inject_service_signal++;
1816 spin_lock(&fi->lock);
1817 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1818
1819 /* We always allow events, track them separately from the sccb ints */
1820 if (fi->srv_signal.ext_params & SCCB_EVENT_PENDING)
1821 set_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1822
1823 /*
1824 * Early versions of the QEMU s390 bios will inject several
1825 * service interrupts after another without handling a
1826 * condition code indicating busy.
1827 * We will silently ignore those superfluous sccb values.
1828 * A future version of QEMU will take care of serialization
1829 * of servc requests
1830 */
1831 if (fi->srv_signal.ext_params & SCCB_MASK)
1832 goto out;
1833 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1834 set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1835out:
1836 spin_unlock(&fi->lock);
1837 kfree(inti);
1838 return 0;
1839}
1840
1841static int __inject_virtio(struct kvm *kvm,
1842 struct kvm_s390_interrupt_info *inti)
1843{
1844 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1845
1846 kvm->stat.inject_virtio++;
1847 spin_lock(&fi->lock);
1848 if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1849 spin_unlock(&fi->lock);
1850 return -EBUSY;
1851 }
1852 fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1853 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1854 set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1855 spin_unlock(&fi->lock);
1856 return 0;
1857}
1858
1859static int __inject_pfault_done(struct kvm *kvm,
1860 struct kvm_s390_interrupt_info *inti)
1861{
1862 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1863
1864 kvm->stat.inject_pfault_done++;
1865 spin_lock(&fi->lock);
1866 if (fi->counters[FIRQ_CNTR_PFAULT] >=
1867 (ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1868 spin_unlock(&fi->lock);
1869 return -EBUSY;
1870 }
1871 fi->counters[FIRQ_CNTR_PFAULT] += 1;
1872 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1873 set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1874 spin_unlock(&fi->lock);
1875 return 0;
1876}
1877
1878#define CR_PENDING_SUBCLASS 28
1879static int __inject_float_mchk(struct kvm *kvm,
1880 struct kvm_s390_interrupt_info *inti)
1881{
1882 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1883
1884 kvm->stat.inject_float_mchk++;
1885 spin_lock(&fi->lock);
1886 fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1887 fi->mchk.mcic |= inti->mchk.mcic;
1888 set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1889 spin_unlock(&fi->lock);
1890 kfree(inti);
1891 return 0;
1892}
1893
1894static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1895{
1896 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1897 struct kvm_s390_float_interrupt *fi;
1898 struct list_head *list;
1899 int isc;
1900
1901 kvm->stat.inject_io++;
1902 isc = int_word_to_isc(inti->io.io_int_word);
1903
1904 /*
1905 * We do not use the lock checking variant as this is just a
1906 * performance optimization and we do not hold the lock here.
1907 * This is ok as the code will pick interrupts from both "lists"
1908 * for delivery.
1909 */
1910 if (gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) {
1911 VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc);
1912 gisa_set_ipm_gisc(gi->origin, isc);
1913 kfree(inti);
1914 return 0;
1915 }
1916
1917 fi = &kvm->arch.float_int;
1918 spin_lock(&fi->lock);
1919 if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1920 spin_unlock(&fi->lock);
1921 return -EBUSY;
1922 }
1923 fi->counters[FIRQ_CNTR_IO] += 1;
1924
1925 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1926 VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1927 else
1928 VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1929 inti->io.subchannel_id >> 8,
1930 inti->io.subchannel_id >> 1 & 0x3,
1931 inti->io.subchannel_nr);
1932 list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1933 list_add_tail(&inti->list, list);
1934 set_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1935 spin_unlock(&fi->lock);
1936 return 0;
1937}
1938
1939/*
1940 * Find a destination VCPU for a floating irq and kick it.
1941 */
1942static void __floating_irq_kick(struct kvm *kvm, u64 type)
1943{
1944 struct kvm_vcpu *dst_vcpu;
1945 int sigcpu, online_vcpus, nr_tries = 0;
1946
1947 online_vcpus = atomic_read(&kvm->online_vcpus);
1948 if (!online_vcpus)
1949 return;
1950
1951 /* find idle VCPUs first, then round robin */
1952 sigcpu = find_first_bit(kvm->arch.idle_mask, online_vcpus);
1953 if (sigcpu == online_vcpus) {
1954 do {
1955 sigcpu = kvm->arch.float_int.next_rr_cpu++;
1956 kvm->arch.float_int.next_rr_cpu %= online_vcpus;
1957 /* avoid endless loops if all vcpus are stopped */
1958 if (nr_tries++ >= online_vcpus)
1959 return;
1960 } while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1961 }
1962 dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1963
1964 /* make the VCPU drop out of the SIE, or wake it up if sleeping */
1965 switch (type) {
1966 case KVM_S390_MCHK:
1967 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT);
1968 break;
1969 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1970 if (!(type & KVM_S390_INT_IO_AI_MASK &&
1971 kvm->arch.gisa_int.origin) ||
1972 kvm_s390_pv_cpu_get_handle(dst_vcpu))
1973 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT);
1974 break;
1975 default:
1976 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT);
1977 break;
1978 }
1979 kvm_s390_vcpu_wakeup(dst_vcpu);
1980}
1981
1982static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1983{
1984 u64 type = READ_ONCE(inti->type);
1985 int rc;
1986
1987 switch (type) {
1988 case KVM_S390_MCHK:
1989 rc = __inject_float_mchk(kvm, inti);
1990 break;
1991 case KVM_S390_INT_VIRTIO:
1992 rc = __inject_virtio(kvm, inti);
1993 break;
1994 case KVM_S390_INT_SERVICE:
1995 rc = __inject_service(kvm, inti);
1996 break;
1997 case KVM_S390_INT_PFAULT_DONE:
1998 rc = __inject_pfault_done(kvm, inti);
1999 break;
2000 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2001 rc = __inject_io(kvm, inti);
2002 break;
2003 default:
2004 rc = -EINVAL;
2005 }
2006 if (rc)
2007 return rc;
2008
2009 __floating_irq_kick(kvm, type);
2010 return 0;
2011}
2012
2013int kvm_s390_inject_vm(struct kvm *kvm,
2014 struct kvm_s390_interrupt *s390int)
2015{
2016 struct kvm_s390_interrupt_info *inti;
2017 int rc;
2018
2019 inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2020 if (!inti)
2021 return -ENOMEM;
2022
2023 inti->type = s390int->type;
2024 switch (inti->type) {
2025 case KVM_S390_INT_VIRTIO:
2026 VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
2027 s390int->parm, s390int->parm64);
2028 inti->ext.ext_params = s390int->parm;
2029 inti->ext.ext_params2 = s390int->parm64;
2030 break;
2031 case KVM_S390_INT_SERVICE:
2032 VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
2033 inti->ext.ext_params = s390int->parm;
2034 break;
2035 case KVM_S390_INT_PFAULT_DONE:
2036 inti->ext.ext_params2 = s390int->parm64;
2037 break;
2038 case KVM_S390_MCHK:
2039 VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
2040 s390int->parm64);
2041 inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
2042 inti->mchk.mcic = s390int->parm64;
2043 break;
2044 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2045 inti->io.subchannel_id = s390int->parm >> 16;
2046 inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
2047 inti->io.io_int_parm = s390int->parm64 >> 32;
2048 inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
2049 break;
2050 default:
2051 kfree(inti);
2052 return -EINVAL;
2053 }
2054 trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
2055 2);
2056
2057 rc = __inject_vm(kvm, inti);
2058 if (rc)
2059 kfree(inti);
2060 return rc;
2061}
2062
2063int kvm_s390_reinject_io_int(struct kvm *kvm,
2064 struct kvm_s390_interrupt_info *inti)
2065{
2066 return __inject_vm(kvm, inti);
2067}
2068
2069int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
2070 struct kvm_s390_irq *irq)
2071{
2072 irq->type = s390int->type;
2073 switch (irq->type) {
2074 case KVM_S390_PROGRAM_INT:
2075 if (s390int->parm & 0xffff0000)
2076 return -EINVAL;
2077 irq->u.pgm.code = s390int->parm;
2078 break;
2079 case KVM_S390_SIGP_SET_PREFIX:
2080 irq->u.prefix.address = s390int->parm;
2081 break;
2082 case KVM_S390_SIGP_STOP:
2083 irq->u.stop.flags = s390int->parm;
2084 break;
2085 case KVM_S390_INT_EXTERNAL_CALL:
2086 if (s390int->parm & 0xffff0000)
2087 return -EINVAL;
2088 irq->u.extcall.code = s390int->parm;
2089 break;
2090 case KVM_S390_INT_EMERGENCY:
2091 if (s390int->parm & 0xffff0000)
2092 return -EINVAL;
2093 irq->u.emerg.code = s390int->parm;
2094 break;
2095 case KVM_S390_MCHK:
2096 irq->u.mchk.mcic = s390int->parm64;
2097 break;
2098 case KVM_S390_INT_PFAULT_INIT:
2099 irq->u.ext.ext_params = s390int->parm;
2100 irq->u.ext.ext_params2 = s390int->parm64;
2101 break;
2102 case KVM_S390_RESTART:
2103 case KVM_S390_INT_CLOCK_COMP:
2104 case KVM_S390_INT_CPU_TIMER:
2105 break;
2106 default:
2107 return -EINVAL;
2108 }
2109 return 0;
2110}
2111
2112int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
2113{
2114 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2115
2116 return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2117}
2118
2119int kvm_s390_is_restart_irq_pending(struct kvm_vcpu *vcpu)
2120{
2121 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2122
2123 return test_bit(IRQ_PEND_RESTART, &li->pending_irqs);
2124}
2125
2126void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
2127{
2128 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2129
2130 spin_lock(&li->lock);
2131 li->irq.stop.flags = 0;
2132 clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2133 spin_unlock(&li->lock);
2134}
2135
2136static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2137{
2138 int rc;
2139
2140 switch (irq->type) {
2141 case KVM_S390_PROGRAM_INT:
2142 rc = __inject_prog(vcpu, irq);
2143 break;
2144 case KVM_S390_SIGP_SET_PREFIX:
2145 rc = __inject_set_prefix(vcpu, irq);
2146 break;
2147 case KVM_S390_SIGP_STOP:
2148 rc = __inject_sigp_stop(vcpu, irq);
2149 break;
2150 case KVM_S390_RESTART:
2151 rc = __inject_sigp_restart(vcpu);
2152 break;
2153 case KVM_S390_INT_CLOCK_COMP:
2154 rc = __inject_ckc(vcpu);
2155 break;
2156 case KVM_S390_INT_CPU_TIMER:
2157 rc = __inject_cpu_timer(vcpu);
2158 break;
2159 case KVM_S390_INT_EXTERNAL_CALL:
2160 rc = __inject_extcall(vcpu, irq);
2161 break;
2162 case KVM_S390_INT_EMERGENCY:
2163 rc = __inject_sigp_emergency(vcpu, irq);
2164 break;
2165 case KVM_S390_MCHK:
2166 rc = __inject_mchk(vcpu, irq);
2167 break;
2168 case KVM_S390_INT_PFAULT_INIT:
2169 rc = __inject_pfault_init(vcpu, irq);
2170 break;
2171 case KVM_S390_INT_VIRTIO:
2172 case KVM_S390_INT_SERVICE:
2173 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2174 default:
2175 rc = -EINVAL;
2176 }
2177
2178 return rc;
2179}
2180
2181int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2182{
2183 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2184 int rc;
2185
2186 spin_lock(&li->lock);
2187 rc = do_inject_vcpu(vcpu, irq);
2188 spin_unlock(&li->lock);
2189 if (!rc)
2190 kvm_s390_vcpu_wakeup(vcpu);
2191 return rc;
2192}
2193
2194static inline void clear_irq_list(struct list_head *_list)
2195{
2196 struct kvm_s390_interrupt_info *inti, *n;
2197
2198 list_for_each_entry_safe(inti, n, _list, list) {
2199 list_del(&inti->list);
2200 kfree(inti);
2201 }
2202}
2203
2204static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
2205 struct kvm_s390_irq *irq)
2206{
2207 irq->type = inti->type;
2208 switch (inti->type) {
2209 case KVM_S390_INT_PFAULT_INIT:
2210 case KVM_S390_INT_PFAULT_DONE:
2211 case KVM_S390_INT_VIRTIO:
2212 irq->u.ext = inti->ext;
2213 break;
2214 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2215 irq->u.io = inti->io;
2216 break;
2217 }
2218}
2219
2220void kvm_s390_clear_float_irqs(struct kvm *kvm)
2221{
2222 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2223 int i;
2224
2225 mutex_lock(&kvm->lock);
2226 if (!kvm_s390_pv_is_protected(kvm))
2227 fi->masked_irqs = 0;
2228 mutex_unlock(&kvm->lock);
2229 spin_lock(&fi->lock);
2230 fi->pending_irqs = 0;
2231 memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
2232 memset(&fi->mchk, 0, sizeof(fi->mchk));
2233 for (i = 0; i < FIRQ_LIST_COUNT; i++)
2234 clear_irq_list(&fi->lists[i]);
2235 for (i = 0; i < FIRQ_MAX_COUNT; i++)
2236 fi->counters[i] = 0;
2237 spin_unlock(&fi->lock);
2238 kvm_s390_gisa_clear(kvm);
2239};
2240
2241static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
2242{
2243 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2244 struct kvm_s390_interrupt_info *inti;
2245 struct kvm_s390_float_interrupt *fi;
2246 struct kvm_s390_irq *buf;
2247 struct kvm_s390_irq *irq;
2248 int max_irqs;
2249 int ret = 0;
2250 int n = 0;
2251 int i;
2252
2253 if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
2254 return -EINVAL;
2255
2256 /*
2257 * We are already using -ENOMEM to signal
2258 * userspace it may retry with a bigger buffer,
2259 * so we need to use something else for this case
2260 */
2261 buf = vzalloc(len);
2262 if (!buf)
2263 return -ENOBUFS;
2264
2265 max_irqs = len / sizeof(struct kvm_s390_irq);
2266
2267 if (gi->origin && gisa_get_ipm(gi->origin)) {
2268 for (i = 0; i <= MAX_ISC; i++) {
2269 if (n == max_irqs) {
2270 /* signal userspace to try again */
2271 ret = -ENOMEM;
2272 goto out_nolock;
2273 }
2274 if (gisa_tac_ipm_gisc(gi->origin, i)) {
2275 irq = (struct kvm_s390_irq *) &buf[n];
2276 irq->type = KVM_S390_INT_IO(1, 0, 0, 0);
2277 irq->u.io.io_int_word = isc_to_int_word(i);
2278 n++;
2279 }
2280 }
2281 }
2282 fi = &kvm->arch.float_int;
2283 spin_lock(&fi->lock);
2284 for (i = 0; i < FIRQ_LIST_COUNT; i++) {
2285 list_for_each_entry(inti, &fi->lists[i], list) {
2286 if (n == max_irqs) {
2287 /* signal userspace to try again */
2288 ret = -ENOMEM;
2289 goto out;
2290 }
2291 inti_to_irq(inti, &buf[n]);
2292 n++;
2293 }
2294 }
2295 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs) ||
2296 test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs)) {
2297 if (n == max_irqs) {
2298 /* signal userspace to try again */
2299 ret = -ENOMEM;
2300 goto out;
2301 }
2302 irq = (struct kvm_s390_irq *) &buf[n];
2303 irq->type = KVM_S390_INT_SERVICE;
2304 irq->u.ext = fi->srv_signal;
2305 n++;
2306 }
2307 if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
2308 if (n == max_irqs) {
2309 /* signal userspace to try again */
2310 ret = -ENOMEM;
2311 goto out;
2312 }
2313 irq = (struct kvm_s390_irq *) &buf[n];
2314 irq->type = KVM_S390_MCHK;
2315 irq->u.mchk = fi->mchk;
2316 n++;
2317}
2318
2319out:
2320 spin_unlock(&fi->lock);
2321out_nolock:
2322 if (!ret && n > 0) {
2323 if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
2324 ret = -EFAULT;
2325 }
2326 vfree(buf);
2327
2328 return ret < 0 ? ret : n;
2329}
2330
2331static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
2332{
2333 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2334 struct kvm_s390_ais_all ais;
2335
2336 if (attr->attr < sizeof(ais))
2337 return -EINVAL;
2338
2339 if (!test_kvm_facility(kvm, 72))
2340 return -EOPNOTSUPP;
2341
2342 mutex_lock(&fi->ais_lock);
2343 ais.simm = fi->simm;
2344 ais.nimm = fi->nimm;
2345 mutex_unlock(&fi->ais_lock);
2346
2347 if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
2348 return -EFAULT;
2349
2350 return 0;
2351}
2352
2353static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2354{
2355 int r;
2356
2357 switch (attr->group) {
2358 case KVM_DEV_FLIC_GET_ALL_IRQS:
2359 r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
2360 attr->attr);
2361 break;
2362 case KVM_DEV_FLIC_AISM_ALL:
2363 r = flic_ais_mode_get_all(dev->kvm, attr);
2364 break;
2365 default:
2366 r = -EINVAL;
2367 }
2368
2369 return r;
2370}
2371
2372static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
2373 u64 addr)
2374{
2375 struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
2376 void *target = NULL;
2377 void __user *source;
2378 u64 size;
2379
2380 if (get_user(inti->type, (u64 __user *)addr))
2381 return -EFAULT;
2382
2383 switch (inti->type) {
2384 case KVM_S390_INT_PFAULT_INIT:
2385 case KVM_S390_INT_PFAULT_DONE:
2386 case KVM_S390_INT_VIRTIO:
2387 case KVM_S390_INT_SERVICE:
2388 target = (void *) &inti->ext;
2389 source = &uptr->u.ext;
2390 size = sizeof(inti->ext);
2391 break;
2392 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2393 target = (void *) &inti->io;
2394 source = &uptr->u.io;
2395 size = sizeof(inti->io);
2396 break;
2397 case KVM_S390_MCHK:
2398 target = (void *) &inti->mchk;
2399 source = &uptr->u.mchk;
2400 size = sizeof(inti->mchk);
2401 break;
2402 default:
2403 return -EINVAL;
2404 }
2405
2406 if (copy_from_user(target, source, size))
2407 return -EFAULT;
2408
2409 return 0;
2410}
2411
2412static int enqueue_floating_irq(struct kvm_device *dev,
2413 struct kvm_device_attr *attr)
2414{
2415 struct kvm_s390_interrupt_info *inti = NULL;
2416 int r = 0;
2417 int len = attr->attr;
2418
2419 if (len % sizeof(struct kvm_s390_irq) != 0)
2420 return -EINVAL;
2421 else if (len > KVM_S390_FLIC_MAX_BUFFER)
2422 return -EINVAL;
2423
2424 while (len >= sizeof(struct kvm_s390_irq)) {
2425 inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2426 if (!inti)
2427 return -ENOMEM;
2428
2429 r = copy_irq_from_user(inti, attr->addr);
2430 if (r) {
2431 kfree(inti);
2432 return r;
2433 }
2434 r = __inject_vm(dev->kvm, inti);
2435 if (r) {
2436 kfree(inti);
2437 return r;
2438 }
2439 len -= sizeof(struct kvm_s390_irq);
2440 attr->addr += sizeof(struct kvm_s390_irq);
2441 }
2442
2443 return r;
2444}
2445
2446static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2447{
2448 if (id >= MAX_S390_IO_ADAPTERS)
2449 return NULL;
2450 id = array_index_nospec(id, MAX_S390_IO_ADAPTERS);
2451 return kvm->arch.adapters[id];
2452}
2453
2454static int register_io_adapter(struct kvm_device *dev,
2455 struct kvm_device_attr *attr)
2456{
2457 struct s390_io_adapter *adapter;
2458 struct kvm_s390_io_adapter adapter_info;
2459
2460 if (copy_from_user(&adapter_info,
2461 (void __user *)attr->addr, sizeof(adapter_info)))
2462 return -EFAULT;
2463
2464 if (adapter_info.id >= MAX_S390_IO_ADAPTERS)
2465 return -EINVAL;
2466
2467 adapter_info.id = array_index_nospec(adapter_info.id,
2468 MAX_S390_IO_ADAPTERS);
2469
2470 if (dev->kvm->arch.adapters[adapter_info.id] != NULL)
2471 return -EINVAL;
2472
2473 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL_ACCOUNT);
2474 if (!adapter)
2475 return -ENOMEM;
2476
2477 adapter->id = adapter_info.id;
2478 adapter->isc = adapter_info.isc;
2479 adapter->maskable = adapter_info.maskable;
2480 adapter->masked = false;
2481 adapter->swap = adapter_info.swap;
2482 adapter->suppressible = (adapter_info.flags) &
2483 KVM_S390_ADAPTER_SUPPRESSIBLE;
2484 dev->kvm->arch.adapters[adapter->id] = adapter;
2485
2486 return 0;
2487}
2488
2489int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2490{
2491 int ret;
2492 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2493
2494 if (!adapter || !adapter->maskable)
2495 return -EINVAL;
2496 ret = adapter->masked;
2497 adapter->masked = masked;
2498 return ret;
2499}
2500
2501void kvm_s390_destroy_adapters(struct kvm *kvm)
2502{
2503 int i;
2504
2505 for (i = 0; i < MAX_S390_IO_ADAPTERS; i++)
2506 kfree(kvm->arch.adapters[i]);
2507}
2508
2509static int modify_io_adapter(struct kvm_device *dev,
2510 struct kvm_device_attr *attr)
2511{
2512 struct kvm_s390_io_adapter_req req;
2513 struct s390_io_adapter *adapter;
2514 int ret;
2515
2516 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2517 return -EFAULT;
2518
2519 adapter = get_io_adapter(dev->kvm, req.id);
2520 if (!adapter)
2521 return -EINVAL;
2522 switch (req.type) {
2523 case KVM_S390_IO_ADAPTER_MASK:
2524 ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2525 if (ret > 0)
2526 ret = 0;
2527 break;
2528 /*
2529 * The following operations are no longer needed and therefore no-ops.
2530 * The gpa to hva translation is done when an IRQ route is set up. The
2531 * set_irq code uses get_user_pages_remote() to do the actual write.
2532 */
2533 case KVM_S390_IO_ADAPTER_MAP:
2534 case KVM_S390_IO_ADAPTER_UNMAP:
2535 ret = 0;
2536 break;
2537 default:
2538 ret = -EINVAL;
2539 }
2540
2541 return ret;
2542}
2543
2544static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2545
2546{
2547 const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2548 u32 schid;
2549
2550 if (attr->flags)
2551 return -EINVAL;
2552 if (attr->attr != sizeof(schid))
2553 return -EINVAL;
2554 if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2555 return -EFAULT;
2556 if (!schid)
2557 return -EINVAL;
2558 kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2559 /*
2560 * If userspace is conforming to the architecture, we can have at most
2561 * one pending I/O interrupt per subchannel, so this is effectively a
2562 * clear all.
2563 */
2564 return 0;
2565}
2566
2567static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2568{
2569 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2570 struct kvm_s390_ais_req req;
2571 int ret = 0;
2572
2573 if (!test_kvm_facility(kvm, 72))
2574 return -EOPNOTSUPP;
2575
2576 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2577 return -EFAULT;
2578
2579 if (req.isc > MAX_ISC)
2580 return -EINVAL;
2581
2582 trace_kvm_s390_modify_ais_mode(req.isc,
2583 (fi->simm & AIS_MODE_MASK(req.isc)) ?
2584 (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2585 2 : KVM_S390_AIS_MODE_SINGLE :
2586 KVM_S390_AIS_MODE_ALL, req.mode);
2587
2588 mutex_lock(&fi->ais_lock);
2589 switch (req.mode) {
2590 case KVM_S390_AIS_MODE_ALL:
2591 fi->simm &= ~AIS_MODE_MASK(req.isc);
2592 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2593 break;
2594 case KVM_S390_AIS_MODE_SINGLE:
2595 fi->simm |= AIS_MODE_MASK(req.isc);
2596 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2597 break;
2598 default:
2599 ret = -EINVAL;
2600 }
2601 mutex_unlock(&fi->ais_lock);
2602
2603 return ret;
2604}
2605
2606static int kvm_s390_inject_airq(struct kvm *kvm,
2607 struct s390_io_adapter *adapter)
2608{
2609 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2610 struct kvm_s390_interrupt s390int = {
2611 .type = KVM_S390_INT_IO(1, 0, 0, 0),
2612 .parm = 0,
2613 .parm64 = isc_to_int_word(adapter->isc),
2614 };
2615 int ret = 0;
2616
2617 if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2618 return kvm_s390_inject_vm(kvm, &s390int);
2619
2620 mutex_lock(&fi->ais_lock);
2621 if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2622 trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2623 goto out;
2624 }
2625
2626 ret = kvm_s390_inject_vm(kvm, &s390int);
2627 if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2628 fi->nimm |= AIS_MODE_MASK(adapter->isc);
2629 trace_kvm_s390_modify_ais_mode(adapter->isc,
2630 KVM_S390_AIS_MODE_SINGLE, 2);
2631 }
2632out:
2633 mutex_unlock(&fi->ais_lock);
2634 return ret;
2635}
2636
2637static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2638{
2639 unsigned int id = attr->attr;
2640 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2641
2642 if (!adapter)
2643 return -EINVAL;
2644
2645 return kvm_s390_inject_airq(kvm, adapter);
2646}
2647
2648static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2649{
2650 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2651 struct kvm_s390_ais_all ais;
2652
2653 if (!test_kvm_facility(kvm, 72))
2654 return -EOPNOTSUPP;
2655
2656 if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2657 return -EFAULT;
2658
2659 mutex_lock(&fi->ais_lock);
2660 fi->simm = ais.simm;
2661 fi->nimm = ais.nimm;
2662 mutex_unlock(&fi->ais_lock);
2663
2664 return 0;
2665}
2666
2667static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2668{
2669 int r = 0;
2670 unsigned long i;
2671 struct kvm_vcpu *vcpu;
2672
2673 switch (attr->group) {
2674 case KVM_DEV_FLIC_ENQUEUE:
2675 r = enqueue_floating_irq(dev, attr);
2676 break;
2677 case KVM_DEV_FLIC_CLEAR_IRQS:
2678 kvm_s390_clear_float_irqs(dev->kvm);
2679 break;
2680 case KVM_DEV_FLIC_APF_ENABLE:
2681 dev->kvm->arch.gmap->pfault_enabled = 1;
2682 break;
2683 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2684 dev->kvm->arch.gmap->pfault_enabled = 0;
2685 /*
2686 * Make sure no async faults are in transition when
2687 * clearing the queues. So we don't need to worry
2688 * about late coming workers.
2689 */
2690 synchronize_srcu(&dev->kvm->srcu);
2691 kvm_for_each_vcpu(i, vcpu, dev->kvm)
2692 kvm_clear_async_pf_completion_queue(vcpu);
2693 break;
2694 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2695 r = register_io_adapter(dev, attr);
2696 break;
2697 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2698 r = modify_io_adapter(dev, attr);
2699 break;
2700 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2701 r = clear_io_irq(dev->kvm, attr);
2702 break;
2703 case KVM_DEV_FLIC_AISM:
2704 r = modify_ais_mode(dev->kvm, attr);
2705 break;
2706 case KVM_DEV_FLIC_AIRQ_INJECT:
2707 r = flic_inject_airq(dev->kvm, attr);
2708 break;
2709 case KVM_DEV_FLIC_AISM_ALL:
2710 r = flic_ais_mode_set_all(dev->kvm, attr);
2711 break;
2712 default:
2713 r = -EINVAL;
2714 }
2715
2716 return r;
2717}
2718
2719static int flic_has_attr(struct kvm_device *dev,
2720 struct kvm_device_attr *attr)
2721{
2722 switch (attr->group) {
2723 case KVM_DEV_FLIC_GET_ALL_IRQS:
2724 case KVM_DEV_FLIC_ENQUEUE:
2725 case KVM_DEV_FLIC_CLEAR_IRQS:
2726 case KVM_DEV_FLIC_APF_ENABLE:
2727 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2728 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2729 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2730 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2731 case KVM_DEV_FLIC_AISM:
2732 case KVM_DEV_FLIC_AIRQ_INJECT:
2733 case KVM_DEV_FLIC_AISM_ALL:
2734 return 0;
2735 }
2736 return -ENXIO;
2737}
2738
2739static int flic_create(struct kvm_device *dev, u32 type)
2740{
2741 if (!dev)
2742 return -EINVAL;
2743 if (dev->kvm->arch.flic)
2744 return -EINVAL;
2745 dev->kvm->arch.flic = dev;
2746 return 0;
2747}
2748
2749static void flic_destroy(struct kvm_device *dev)
2750{
2751 dev->kvm->arch.flic = NULL;
2752 kfree(dev);
2753}
2754
2755/* s390 floating irq controller (flic) */
2756struct kvm_device_ops kvm_flic_ops = {
2757 .name = "kvm-flic",
2758 .get_attr = flic_get_attr,
2759 .set_attr = flic_set_attr,
2760 .has_attr = flic_has_attr,
2761 .create = flic_create,
2762 .destroy = flic_destroy,
2763};
2764
2765static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2766{
2767 unsigned long bit;
2768
2769 bit = bit_nr + (addr % PAGE_SIZE) * 8;
2770
2771 return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2772}
2773
2774static struct page *get_map_page(struct kvm *kvm, u64 uaddr)
2775{
2776 struct page *page = NULL;
2777
2778 mmap_read_lock(kvm->mm);
2779 get_user_pages_remote(kvm->mm, uaddr, 1, FOLL_WRITE,
2780 &page, NULL, NULL);
2781 mmap_read_unlock(kvm->mm);
2782 return page;
2783}
2784
2785static int adapter_indicators_set(struct kvm *kvm,
2786 struct s390_io_adapter *adapter,
2787 struct kvm_s390_adapter_int *adapter_int)
2788{
2789 unsigned long bit;
2790 int summary_set, idx;
2791 struct page *ind_page, *summary_page;
2792 void *map;
2793
2794 ind_page = get_map_page(kvm, adapter_int->ind_addr);
2795 if (!ind_page)
2796 return -1;
2797 summary_page = get_map_page(kvm, adapter_int->summary_addr);
2798 if (!summary_page) {
2799 put_page(ind_page);
2800 return -1;
2801 }
2802
2803 idx = srcu_read_lock(&kvm->srcu);
2804 map = page_address(ind_page);
2805 bit = get_ind_bit(adapter_int->ind_addr,
2806 adapter_int->ind_offset, adapter->swap);
2807 set_bit(bit, map);
2808 mark_page_dirty(kvm, adapter_int->ind_addr >> PAGE_SHIFT);
2809 set_page_dirty_lock(ind_page);
2810 map = page_address(summary_page);
2811 bit = get_ind_bit(adapter_int->summary_addr,
2812 adapter_int->summary_offset, adapter->swap);
2813 summary_set = test_and_set_bit(bit, map);
2814 mark_page_dirty(kvm, adapter_int->summary_addr >> PAGE_SHIFT);
2815 set_page_dirty_lock(summary_page);
2816 srcu_read_unlock(&kvm->srcu, idx);
2817
2818 put_page(ind_page);
2819 put_page(summary_page);
2820 return summary_set ? 0 : 1;
2821}
2822
2823/*
2824 * < 0 - not injected due to error
2825 * = 0 - coalesced, summary indicator already active
2826 * > 0 - injected interrupt
2827 */
2828static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2829 struct kvm *kvm, int irq_source_id, int level,
2830 bool line_status)
2831{
2832 int ret;
2833 struct s390_io_adapter *adapter;
2834
2835 /* We're only interested in the 0->1 transition. */
2836 if (!level)
2837 return 0;
2838 adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2839 if (!adapter)
2840 return -1;
2841 ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2842 if ((ret > 0) && !adapter->masked) {
2843 ret = kvm_s390_inject_airq(kvm, adapter);
2844 if (ret == 0)
2845 ret = 1;
2846 }
2847 return ret;
2848}
2849
2850/*
2851 * Inject the machine check to the guest.
2852 */
2853void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2854 struct mcck_volatile_info *mcck_info)
2855{
2856 struct kvm_s390_interrupt_info inti;
2857 struct kvm_s390_irq irq;
2858 struct kvm_s390_mchk_info *mchk;
2859 union mci mci;
2860 __u64 cr14 = 0; /* upper bits are not used */
2861 int rc;
2862
2863 mci.val = mcck_info->mcic;
2864 if (mci.sr)
2865 cr14 |= CR14_RECOVERY_SUBMASK;
2866 if (mci.dg)
2867 cr14 |= CR14_DEGRADATION_SUBMASK;
2868 if (mci.w)
2869 cr14 |= CR14_WARNING_SUBMASK;
2870
2871 mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2872 mchk->cr14 = cr14;
2873 mchk->mcic = mcck_info->mcic;
2874 mchk->ext_damage_code = mcck_info->ext_damage_code;
2875 mchk->failing_storage_address = mcck_info->failing_storage_address;
2876 if (mci.ck) {
2877 /* Inject the floating machine check */
2878 inti.type = KVM_S390_MCHK;
2879 rc = __inject_vm(vcpu->kvm, &inti);
2880 } else {
2881 /* Inject the machine check to specified vcpu */
2882 irq.type = KVM_S390_MCHK;
2883 rc = kvm_s390_inject_vcpu(vcpu, &irq);
2884 }
2885 WARN_ON_ONCE(rc);
2886}
2887
2888int kvm_set_routing_entry(struct kvm *kvm,
2889 struct kvm_kernel_irq_routing_entry *e,
2890 const struct kvm_irq_routing_entry *ue)
2891{
2892 u64 uaddr;
2893
2894 switch (ue->type) {
2895 /* we store the userspace addresses instead of the guest addresses */
2896 case KVM_IRQ_ROUTING_S390_ADAPTER:
2897 e->set = set_adapter_int;
2898 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.summary_addr);
2899 if (uaddr == -EFAULT)
2900 return -EFAULT;
2901 e->adapter.summary_addr = uaddr;
2902 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.ind_addr);
2903 if (uaddr == -EFAULT)
2904 return -EFAULT;
2905 e->adapter.ind_addr = uaddr;
2906 e->adapter.summary_offset = ue->u.adapter.summary_offset;
2907 e->adapter.ind_offset = ue->u.adapter.ind_offset;
2908 e->adapter.adapter_id = ue->u.adapter.adapter_id;
2909 return 0;
2910 default:
2911 return -EINVAL;
2912 }
2913}
2914
2915int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2916 int irq_source_id, int level, bool line_status)
2917{
2918 return -EINVAL;
2919}
2920
2921int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2922{
2923 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2924 struct kvm_s390_irq *buf;
2925 int r = 0;
2926 int n;
2927
2928 buf = vmalloc(len);
2929 if (!buf)
2930 return -ENOMEM;
2931
2932 if (copy_from_user((void *) buf, irqstate, len)) {
2933 r = -EFAULT;
2934 goto out_free;
2935 }
2936
2937 /*
2938 * Don't allow setting the interrupt state
2939 * when there are already interrupts pending
2940 */
2941 spin_lock(&li->lock);
2942 if (li->pending_irqs) {
2943 r = -EBUSY;
2944 goto out_unlock;
2945 }
2946
2947 for (n = 0; n < len / sizeof(*buf); n++) {
2948 r = do_inject_vcpu(vcpu, &buf[n]);
2949 if (r)
2950 break;
2951 }
2952
2953out_unlock:
2954 spin_unlock(&li->lock);
2955out_free:
2956 vfree(buf);
2957
2958 return r;
2959}
2960
2961static void store_local_irq(struct kvm_s390_local_interrupt *li,
2962 struct kvm_s390_irq *irq,
2963 unsigned long irq_type)
2964{
2965 switch (irq_type) {
2966 case IRQ_PEND_MCHK_EX:
2967 case IRQ_PEND_MCHK_REP:
2968 irq->type = KVM_S390_MCHK;
2969 irq->u.mchk = li->irq.mchk;
2970 break;
2971 case IRQ_PEND_PROG:
2972 irq->type = KVM_S390_PROGRAM_INT;
2973 irq->u.pgm = li->irq.pgm;
2974 break;
2975 case IRQ_PEND_PFAULT_INIT:
2976 irq->type = KVM_S390_INT_PFAULT_INIT;
2977 irq->u.ext = li->irq.ext;
2978 break;
2979 case IRQ_PEND_EXT_EXTERNAL:
2980 irq->type = KVM_S390_INT_EXTERNAL_CALL;
2981 irq->u.extcall = li->irq.extcall;
2982 break;
2983 case IRQ_PEND_EXT_CLOCK_COMP:
2984 irq->type = KVM_S390_INT_CLOCK_COMP;
2985 break;
2986 case IRQ_PEND_EXT_CPU_TIMER:
2987 irq->type = KVM_S390_INT_CPU_TIMER;
2988 break;
2989 case IRQ_PEND_SIGP_STOP:
2990 irq->type = KVM_S390_SIGP_STOP;
2991 irq->u.stop = li->irq.stop;
2992 break;
2993 case IRQ_PEND_RESTART:
2994 irq->type = KVM_S390_RESTART;
2995 break;
2996 case IRQ_PEND_SET_PREFIX:
2997 irq->type = KVM_S390_SIGP_SET_PREFIX;
2998 irq->u.prefix = li->irq.prefix;
2999 break;
3000 }
3001}
3002
3003int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
3004{
3005 int scn;
3006 DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
3007 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
3008 unsigned long pending_irqs;
3009 struct kvm_s390_irq irq;
3010 unsigned long irq_type;
3011 int cpuaddr;
3012 int n = 0;
3013
3014 spin_lock(&li->lock);
3015 pending_irqs = li->pending_irqs;
3016 memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
3017 sizeof(sigp_emerg_pending));
3018 spin_unlock(&li->lock);
3019
3020 for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
3021 memset(&irq, 0, sizeof(irq));
3022 if (irq_type == IRQ_PEND_EXT_EMERGENCY)
3023 continue;
3024 if (n + sizeof(irq) > len)
3025 return -ENOBUFS;
3026 store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
3027 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3028 return -EFAULT;
3029 n += sizeof(irq);
3030 }
3031
3032 if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
3033 for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
3034 memset(&irq, 0, sizeof(irq));
3035 if (n + sizeof(irq) > len)
3036 return -ENOBUFS;
3037 irq.type = KVM_S390_INT_EMERGENCY;
3038 irq.u.emerg.code = cpuaddr;
3039 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3040 return -EFAULT;
3041 n += sizeof(irq);
3042 }
3043 }
3044
3045 if (sca_ext_call_pending(vcpu, &scn)) {
3046 if (n + sizeof(irq) > len)
3047 return -ENOBUFS;
3048 memset(&irq, 0, sizeof(irq));
3049 irq.type = KVM_S390_INT_EXTERNAL_CALL;
3050 irq.u.extcall.code = scn;
3051 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3052 return -EFAULT;
3053 n += sizeof(irq);
3054 }
3055
3056 return n;
3057}
3058
3059static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
3060{
3061 int vcpu_idx, online_vcpus = atomic_read(&kvm->online_vcpus);
3062 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3063 struct kvm_vcpu *vcpu;
3064 u8 vcpu_isc_mask;
3065
3066 for_each_set_bit(vcpu_idx, kvm->arch.idle_mask, online_vcpus) {
3067 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
3068 if (psw_ioint_disabled(vcpu))
3069 continue;
3070 vcpu_isc_mask = (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
3071 if (deliverable_mask & vcpu_isc_mask) {
3072 /* lately kicked but not yet running */
3073 if (test_and_set_bit(vcpu_idx, gi->kicked_mask))
3074 return;
3075 kvm_s390_vcpu_wakeup(vcpu);
3076 return;
3077 }
3078 }
3079}
3080
3081static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3082{
3083 struct kvm_s390_gisa_interrupt *gi =
3084 container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3085 struct kvm *kvm =
3086 container_of(gi->origin, struct sie_page2, gisa)->kvm;
3087 u8 pending_mask;
3088
3089 pending_mask = gisa_get_ipm_or_restore_iam(gi);
3090 if (pending_mask) {
3091 __airqs_kick_single_vcpu(kvm, pending_mask);
3092 hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3093 return HRTIMER_RESTART;
3094 }
3095
3096 return HRTIMER_NORESTART;
3097}
3098
3099#define NULL_GISA_ADDR 0x00000000UL
3100#define NONE_GISA_ADDR 0x00000001UL
3101#define GISA_ADDR_MASK 0xfffff000UL
3102
3103static void process_gib_alert_list(void)
3104{
3105 struct kvm_s390_gisa_interrupt *gi;
3106 struct kvm_s390_gisa *gisa;
3107 struct kvm *kvm;
3108 u32 final, origin = 0UL;
3109
3110 do {
3111 /*
3112 * If the NONE_GISA_ADDR is still stored in the alert list
3113 * origin, we will leave the outer loop. No further GISA has
3114 * been added to the alert list by millicode while processing
3115 * the current alert list.
3116 */
3117 final = (origin & NONE_GISA_ADDR);
3118 /*
3119 * Cut off the alert list and store the NONE_GISA_ADDR in the
3120 * alert list origin to avoid further GAL interruptions.
3121 * A new alert list can be build up by millicode in parallel
3122 * for guests not in the yet cut-off alert list. When in the
3123 * final loop, store the NULL_GISA_ADDR instead. This will re-
3124 * enable GAL interruptions on the host again.
3125 */
3126 origin = xchg(&gib->alert_list_origin,
3127 (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3128 /*
3129 * Loop through the just cut-off alert list and start the
3130 * gisa timers to kick idle vcpus to consume the pending
3131 * interruptions asap.
3132 */
3133 while (origin & GISA_ADDR_MASK) {
3134 gisa = (struct kvm_s390_gisa *)(u64)origin;
3135 origin = gisa->next_alert;
3136 gisa->next_alert = (u32)(u64)gisa;
3137 kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3138 gi = &kvm->arch.gisa_int;
3139 if (hrtimer_active(&gi->timer))
3140 hrtimer_cancel(&gi->timer);
3141 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3142 }
3143 } while (!final);
3144
3145}
3146
3147void kvm_s390_gisa_clear(struct kvm *kvm)
3148{
3149 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3150
3151 if (!gi->origin)
3152 return;
3153 gisa_clear_ipm(gi->origin);
3154 VM_EVENT(kvm, 3, "gisa 0x%pK cleared", gi->origin);
3155}
3156
3157void kvm_s390_gisa_init(struct kvm *kvm)
3158{
3159 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3160
3161 if (!css_general_characteristics.aiv)
3162 return;
3163 gi->origin = &kvm->arch.sie_page2->gisa;
3164 gi->alert.mask = 0;
3165 spin_lock_init(&gi->alert.ref_lock);
3166 gi->expires = 50 * 1000; /* 50 usec */
3167 hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3168 gi->timer.function = gisa_vcpu_kicker;
3169 memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3170 gi->origin->next_alert = (u32)(u64)gi->origin;
3171 VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin);
3172}
3173
3174void kvm_s390_gisa_enable(struct kvm *kvm)
3175{
3176 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3177 struct kvm_vcpu *vcpu;
3178 unsigned long i;
3179 u32 gisa_desc;
3180
3181 if (gi->origin)
3182 return;
3183 kvm_s390_gisa_init(kvm);
3184 gisa_desc = kvm_s390_get_gisa_desc(kvm);
3185 if (!gisa_desc)
3186 return;
3187 kvm_for_each_vcpu(i, vcpu, kvm) {
3188 mutex_lock(&vcpu->mutex);
3189 vcpu->arch.sie_block->gd = gisa_desc;
3190 vcpu->arch.sie_block->eca |= ECA_AIV;
3191 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3192 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3193 mutex_unlock(&vcpu->mutex);
3194 }
3195}
3196
3197void kvm_s390_gisa_destroy(struct kvm *kvm)
3198{
3199 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3200 struct kvm_s390_gisa *gisa = gi->origin;
3201
3202 if (!gi->origin)
3203 return;
3204 if (gi->alert.mask)
3205 KVM_EVENT(3, "vm 0x%pK has unexpected iam 0x%02x",
3206 kvm, gi->alert.mask);
3207 while (gisa_in_alert_list(gi->origin))
3208 cpu_relax();
3209 hrtimer_cancel(&gi->timer);
3210 gi->origin = NULL;
3211 VM_EVENT(kvm, 3, "gisa 0x%pK destroyed", gisa);
3212}
3213
3214void kvm_s390_gisa_disable(struct kvm *kvm)
3215{
3216 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3217 struct kvm_vcpu *vcpu;
3218 unsigned long i;
3219
3220 if (!gi->origin)
3221 return;
3222 kvm_for_each_vcpu(i, vcpu, kvm) {
3223 mutex_lock(&vcpu->mutex);
3224 vcpu->arch.sie_block->eca &= ~ECA_AIV;
3225 vcpu->arch.sie_block->gd = 0U;
3226 mutex_unlock(&vcpu->mutex);
3227 VCPU_EVENT(vcpu, 3, "AIV disabled for cpu %03u", vcpu->vcpu_id);
3228 }
3229 kvm_s390_gisa_destroy(kvm);
3230}
3231
3232/**
3233 * kvm_s390_gisc_register - register a guest ISC
3234 *
3235 * @kvm: the kernel vm to work with
3236 * @gisc: the guest interruption sub class to register
3237 *
3238 * The function extends the vm specific alert mask to use.
3239 * The effective IAM mask in the GISA is updated as well
3240 * in case the GISA is not part of the GIB alert list.
3241 * It will be updated latest when the IAM gets restored
3242 * by gisa_get_ipm_or_restore_iam().
3243 *
3244 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3245 * has registered with the channel subsystem.
3246 * -ENODEV in case the vm uses no GISA
3247 * -ERANGE in case the guest ISC is invalid
3248 */
3249int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3250{
3251 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3252
3253 if (!gi->origin)
3254 return -ENODEV;
3255 if (gisc > MAX_ISC)
3256 return -ERANGE;
3257
3258 spin_lock(&gi->alert.ref_lock);
3259 gi->alert.ref_count[gisc]++;
3260 if (gi->alert.ref_count[gisc] == 1) {
3261 gi->alert.mask |= 0x80 >> gisc;
3262 gisa_set_iam(gi->origin, gi->alert.mask);
3263 }
3264 spin_unlock(&gi->alert.ref_lock);
3265
3266 return gib->nisc;
3267}
3268EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3269
3270/**
3271 * kvm_s390_gisc_unregister - unregister a guest ISC
3272 *
3273 * @kvm: the kernel vm to work with
3274 * @gisc: the guest interruption sub class to register
3275 *
3276 * The function reduces the vm specific alert mask to use.
3277 * The effective IAM mask in the GISA is updated as well
3278 * in case the GISA is not part of the GIB alert list.
3279 * It will be updated latest when the IAM gets restored
3280 * by gisa_get_ipm_or_restore_iam().
3281 *
3282 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3283 * has registered with the channel subsystem.
3284 * -ENODEV in case the vm uses no GISA
3285 * -ERANGE in case the guest ISC is invalid
3286 * -EINVAL in case the guest ISC is not registered
3287 */
3288int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3289{
3290 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3291 int rc = 0;
3292
3293 if (!gi->origin)
3294 return -ENODEV;
3295 if (gisc > MAX_ISC)
3296 return -ERANGE;
3297
3298 spin_lock(&gi->alert.ref_lock);
3299 if (gi->alert.ref_count[gisc] == 0) {
3300 rc = -EINVAL;
3301 goto out;
3302 }
3303 gi->alert.ref_count[gisc]--;
3304 if (gi->alert.ref_count[gisc] == 0) {
3305 gi->alert.mask &= ~(0x80 >> gisc);
3306 gisa_set_iam(gi->origin, gi->alert.mask);
3307 }
3308out:
3309 spin_unlock(&gi->alert.ref_lock);
3310
3311 return rc;
3312}
3313EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3314
3315static void aen_host_forward(unsigned long si)
3316{
3317 struct kvm_s390_gisa_interrupt *gi;
3318 struct zpci_gaite *gaite;
3319 struct kvm *kvm;
3320
3321 gaite = (struct zpci_gaite *)aift->gait +
3322 (si * sizeof(struct zpci_gaite));
3323 if (gaite->count == 0)
3324 return;
3325 if (gaite->aisb != 0)
3326 set_bit_inv(gaite->aisbo, phys_to_virt(gaite->aisb));
3327
3328 kvm = kvm_s390_pci_si_to_kvm(aift, si);
3329 if (!kvm)
3330 return;
3331 gi = &kvm->arch.gisa_int;
3332
3333 if (!(gi->origin->g1.simm & AIS_MODE_MASK(gaite->gisc)) ||
3334 !(gi->origin->g1.nimm & AIS_MODE_MASK(gaite->gisc))) {
3335 gisa_set_ipm_gisc(gi->origin, gaite->gisc);
3336 if (hrtimer_active(&gi->timer))
3337 hrtimer_cancel(&gi->timer);
3338 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3339 kvm->stat.aen_forward++;
3340 }
3341}
3342
3343static void aen_process_gait(u8 isc)
3344{
3345 bool found = false, first = true;
3346 union zpci_sic_iib iib = {{0}};
3347 unsigned long si, flags;
3348
3349 spin_lock_irqsave(&aift->gait_lock, flags);
3350
3351 if (!aift->gait) {
3352 spin_unlock_irqrestore(&aift->gait_lock, flags);
3353 return;
3354 }
3355
3356 for (si = 0;;) {
3357 /* Scan adapter summary indicator bit vector */
3358 si = airq_iv_scan(aift->sbv, si, airq_iv_end(aift->sbv));
3359 if (si == -1UL) {
3360 if (first || found) {
3361 /* Re-enable interrupts. */
3362 zpci_set_irq_ctrl(SIC_IRQ_MODE_SINGLE, isc,
3363 &iib);
3364 first = found = false;
3365 } else {
3366 /* Interrupts on and all bits processed */
3367 break;
3368 }
3369 found = false;
3370 si = 0;
3371 /* Scan again after re-enabling interrupts */
3372 continue;
3373 }
3374 found = true;
3375 aen_host_forward(si);
3376 }
3377
3378 spin_unlock_irqrestore(&aift->gait_lock, flags);
3379}
3380
3381static void gib_alert_irq_handler(struct airq_struct *airq,
3382 struct tpi_info *tpi_info)
3383{
3384 struct tpi_adapter_info *info = (struct tpi_adapter_info *)tpi_info;
3385
3386 inc_irq_stat(IRQIO_GAL);
3387
3388 if ((info->forward || info->error) &&
3389 IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3390 aen_process_gait(info->isc);
3391 if (info->aism != 0)
3392 process_gib_alert_list();
3393 } else {
3394 process_gib_alert_list();
3395 }
3396}
3397
3398static struct airq_struct gib_alert_irq = {
3399 .handler = gib_alert_irq_handler,
3400 .lsi_ptr = &gib_alert_irq.lsi_mask,
3401};
3402
3403void kvm_s390_gib_destroy(void)
3404{
3405 if (!gib)
3406 return;
3407 if (kvm_s390_pci_interp_allowed() && aift) {
3408 mutex_lock(&aift->aift_lock);
3409 kvm_s390_pci_aen_exit();
3410 mutex_unlock(&aift->aift_lock);
3411 }
3412 chsc_sgib(0);
3413 unregister_adapter_interrupt(&gib_alert_irq);
3414 free_page((unsigned long)gib);
3415 gib = NULL;
3416}
3417
3418int kvm_s390_gib_init(u8 nisc)
3419{
3420 int rc = 0;
3421
3422 if (!css_general_characteristics.aiv) {
3423 KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3424 goto out;
3425 }
3426
3427 gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3428 if (!gib) {
3429 rc = -ENOMEM;
3430 goto out;
3431 }
3432
3433 gib_alert_irq.isc = nisc;
3434 if (register_adapter_interrupt(&gib_alert_irq)) {
3435 pr_err("Registering the GIB alert interruption handler failed\n");
3436 rc = -EIO;
3437 goto out_free_gib;
3438 }
3439
3440 gib->nisc = nisc;
3441 if (chsc_sgib((u32)(u64)gib)) {
3442 pr_err("Associating the GIB with the AIV facility failed\n");
3443 free_page((unsigned long)gib);
3444 gib = NULL;
3445 rc = -EIO;
3446 goto out_unreg_gal;
3447 }
3448
3449 if (kvm_s390_pci_interp_allowed()) {
3450 if (kvm_s390_pci_aen_init(nisc)) {
3451 pr_err("Initializing AEN for PCI failed\n");
3452 rc = -EIO;
3453 goto out_unreg_gal;
3454 }
3455 }
3456
3457 KVM_EVENT(3, "gib 0x%pK (nisc=%d) initialized", gib, gib->nisc);
3458 goto out;
3459
3460out_unreg_gal:
3461 unregister_adapter_interrupt(&gib_alert_irq);
3462out_free_gib:
3463 free_page((unsigned long)gib);
3464 gib = NULL;
3465out:
3466 return rc;
3467}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * handling kvm guest interrupts
4 *
5 * Copyright IBM Corp. 2008, 2015
6 *
7 * Author(s): Carsten Otte <cotte@de.ibm.com>
8 */
9
10#define KMSG_COMPONENT "kvm-s390"
11#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
12
13#include <linux/interrupt.h>
14#include <linux/kvm_host.h>
15#include <linux/hrtimer.h>
16#include <linux/mmu_context.h>
17#include <linux/nospec.h>
18#include <linux/signal.h>
19#include <linux/slab.h>
20#include <linux/bitmap.h>
21#include <linux/vmalloc.h>
22#include <asm/asm-offsets.h>
23#include <asm/dis.h>
24#include <linux/uaccess.h>
25#include <asm/sclp.h>
26#include <asm/isc.h>
27#include <asm/gmap.h>
28#include <asm/switch_to.h>
29#include <asm/nmi.h>
30#include <asm/airq.h>
31#include "kvm-s390.h"
32#include "gaccess.h"
33#include "trace-s390.h"
34
35#define PFAULT_INIT 0x0600
36#define PFAULT_DONE 0x0680
37#define VIRTIO_PARAM 0x0d00
38
39static struct kvm_s390_gib *gib;
40
41/* handle external calls via sigp interpretation facility */
42static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
43{
44 int c, scn;
45
46 if (!kvm_s390_test_cpuflags(vcpu, CPUSTAT_ECALL_PEND))
47 return 0;
48
49 BUG_ON(!kvm_s390_use_sca_entries());
50 read_lock(&vcpu->kvm->arch.sca_lock);
51 if (vcpu->kvm->arch.use_esca) {
52 struct esca_block *sca = vcpu->kvm->arch.sca;
53 union esca_sigp_ctrl sigp_ctrl =
54 sca->cpu[vcpu->vcpu_id].sigp_ctrl;
55
56 c = sigp_ctrl.c;
57 scn = sigp_ctrl.scn;
58 } else {
59 struct bsca_block *sca = vcpu->kvm->arch.sca;
60 union bsca_sigp_ctrl sigp_ctrl =
61 sca->cpu[vcpu->vcpu_id].sigp_ctrl;
62
63 c = sigp_ctrl.c;
64 scn = sigp_ctrl.scn;
65 }
66 read_unlock(&vcpu->kvm->arch.sca_lock);
67
68 if (src_id)
69 *src_id = scn;
70
71 return c;
72}
73
74static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id)
75{
76 int expect, rc;
77
78 BUG_ON(!kvm_s390_use_sca_entries());
79 read_lock(&vcpu->kvm->arch.sca_lock);
80 if (vcpu->kvm->arch.use_esca) {
81 struct esca_block *sca = vcpu->kvm->arch.sca;
82 union esca_sigp_ctrl *sigp_ctrl =
83 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
84 union esca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
85
86 new_val.scn = src_id;
87 new_val.c = 1;
88 old_val.c = 0;
89
90 expect = old_val.value;
91 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
92 } else {
93 struct bsca_block *sca = vcpu->kvm->arch.sca;
94 union bsca_sigp_ctrl *sigp_ctrl =
95 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
96 union bsca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
97
98 new_val.scn = src_id;
99 new_val.c = 1;
100 old_val.c = 0;
101
102 expect = old_val.value;
103 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
104 }
105 read_unlock(&vcpu->kvm->arch.sca_lock);
106
107 if (rc != expect) {
108 /* another external call is pending */
109 return -EBUSY;
110 }
111 kvm_s390_set_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
112 return 0;
113}
114
115static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
116{
117 int rc, expect;
118
119 if (!kvm_s390_use_sca_entries())
120 return;
121 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
122 read_lock(&vcpu->kvm->arch.sca_lock);
123 if (vcpu->kvm->arch.use_esca) {
124 struct esca_block *sca = vcpu->kvm->arch.sca;
125 union esca_sigp_ctrl *sigp_ctrl =
126 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
127 union esca_sigp_ctrl old = *sigp_ctrl;
128
129 expect = old.value;
130 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
131 } else {
132 struct bsca_block *sca = vcpu->kvm->arch.sca;
133 union bsca_sigp_ctrl *sigp_ctrl =
134 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
135 union bsca_sigp_ctrl old = *sigp_ctrl;
136
137 expect = old.value;
138 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
139 }
140 read_unlock(&vcpu->kvm->arch.sca_lock);
141 WARN_ON(rc != expect); /* cannot clear? */
142}
143
144int psw_extint_disabled(struct kvm_vcpu *vcpu)
145{
146 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
147}
148
149static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
150{
151 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
152}
153
154static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
155{
156 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
157}
158
159static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
160{
161 return psw_extint_disabled(vcpu) &&
162 psw_ioint_disabled(vcpu) &&
163 psw_mchk_disabled(vcpu);
164}
165
166static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
167{
168 if (psw_extint_disabled(vcpu) ||
169 !(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
170 return 0;
171 if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
172 /* No timer interrupts when single stepping */
173 return 0;
174 return 1;
175}
176
177static int ckc_irq_pending(struct kvm_vcpu *vcpu)
178{
179 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
180 const u64 ckc = vcpu->arch.sie_block->ckc;
181
182 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
183 if ((s64)ckc >= (s64)now)
184 return 0;
185 } else if (ckc >= now) {
186 return 0;
187 }
188 return ckc_interrupts_enabled(vcpu);
189}
190
191static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
192{
193 return !psw_extint_disabled(vcpu) &&
194 (vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK);
195}
196
197static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
198{
199 if (!cpu_timer_interrupts_enabled(vcpu))
200 return 0;
201 return kvm_s390_get_cpu_timer(vcpu) >> 63;
202}
203
204static uint64_t isc_to_isc_bits(int isc)
205{
206 return (0x80 >> isc) << 24;
207}
208
209static inline u32 isc_to_int_word(u8 isc)
210{
211 return ((u32)isc << 27) | 0x80000000;
212}
213
214static inline u8 int_word_to_isc(u32 int_word)
215{
216 return (int_word & 0x38000000) >> 27;
217}
218
219/*
220 * To use atomic bitmap functions, we have to provide a bitmap address
221 * that is u64 aligned. However, the ipm might be u32 aligned.
222 * Therefore, we logically start the bitmap at the very beginning of the
223 * struct and fixup the bit number.
224 */
225#define IPM_BIT_OFFSET (offsetof(struct kvm_s390_gisa, ipm) * BITS_PER_BYTE)
226
227/**
228 * gisa_set_iam - change the GISA interruption alert mask
229 *
230 * @gisa: gisa to operate on
231 * @iam: new IAM value to use
232 *
233 * Change the IAM atomically with the next alert address and the IPM
234 * of the GISA if the GISA is not part of the GIB alert list. All three
235 * fields are located in the first long word of the GISA.
236 *
237 * Returns: 0 on success
238 * -EBUSY in case the gisa is part of the alert list
239 */
240static inline int gisa_set_iam(struct kvm_s390_gisa *gisa, u8 iam)
241{
242 u64 word, _word;
243
244 do {
245 word = READ_ONCE(gisa->u64.word[0]);
246 if ((u64)gisa != word >> 32)
247 return -EBUSY;
248 _word = (word & ~0xffUL) | iam;
249 } while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
250
251 return 0;
252}
253
254/**
255 * gisa_clear_ipm - clear the GISA interruption pending mask
256 *
257 * @gisa: gisa to operate on
258 *
259 * Clear the IPM atomically with the next alert address and the IAM
260 * of the GISA unconditionally. All three fields are located in the
261 * first long word of the GISA.
262 */
263static inline void gisa_clear_ipm(struct kvm_s390_gisa *gisa)
264{
265 u64 word, _word;
266
267 do {
268 word = READ_ONCE(gisa->u64.word[0]);
269 _word = word & ~(0xffUL << 24);
270 } while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
271}
272
273/**
274 * gisa_get_ipm_or_restore_iam - return IPM or restore GISA IAM
275 *
276 * @gi: gisa interrupt struct to work on
277 *
278 * Atomically restores the interruption alert mask if none of the
279 * relevant ISCs are pending and return the IPM.
280 *
281 * Returns: the relevant pending ISCs
282 */
283static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi)
284{
285 u8 pending_mask, alert_mask;
286 u64 word, _word;
287
288 do {
289 word = READ_ONCE(gi->origin->u64.word[0]);
290 alert_mask = READ_ONCE(gi->alert.mask);
291 pending_mask = (u8)(word >> 24) & alert_mask;
292 if (pending_mask)
293 return pending_mask;
294 _word = (word & ~0xffUL) | alert_mask;
295 } while (cmpxchg(&gi->origin->u64.word[0], word, _word) != word);
296
297 return 0;
298}
299
300static inline int gisa_in_alert_list(struct kvm_s390_gisa *gisa)
301{
302 return READ_ONCE(gisa->next_alert) != (u32)(u64)gisa;
303}
304
305static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
306{
307 set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
308}
309
310static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa)
311{
312 return READ_ONCE(gisa->ipm);
313}
314
315static inline void gisa_clear_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
316{
317 clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
318}
319
320static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
321{
322 return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
323}
324
325static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu)
326{
327 return vcpu->kvm->arch.float_int.pending_irqs |
328 vcpu->arch.local_int.pending_irqs;
329}
330
331static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
332{
333 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
334 unsigned long pending_mask;
335
336 pending_mask = pending_irqs_no_gisa(vcpu);
337 if (gi->origin)
338 pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7;
339 return pending_mask;
340}
341
342static inline int isc_to_irq_type(unsigned long isc)
343{
344 return IRQ_PEND_IO_ISC_0 - isc;
345}
346
347static inline int irq_type_to_isc(unsigned long irq_type)
348{
349 return IRQ_PEND_IO_ISC_0 - irq_type;
350}
351
352static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
353 unsigned long active_mask)
354{
355 int i;
356
357 for (i = 0; i <= MAX_ISC; i++)
358 if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
359 active_mask &= ~(1UL << (isc_to_irq_type(i)));
360
361 return active_mask;
362}
363
364static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
365{
366 unsigned long active_mask;
367
368 active_mask = pending_irqs(vcpu);
369 if (!active_mask)
370 return 0;
371
372 if (psw_extint_disabled(vcpu))
373 active_mask &= ~IRQ_PEND_EXT_MASK;
374 if (psw_ioint_disabled(vcpu))
375 active_mask &= ~IRQ_PEND_IO_MASK;
376 else
377 active_mask = disable_iscs(vcpu, active_mask);
378 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
379 __clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
380 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK))
381 __clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
382 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
383 __clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
384 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK))
385 __clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
386 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK))
387 __clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
388 if (psw_mchk_disabled(vcpu))
389 active_mask &= ~IRQ_PEND_MCHK_MASK;
390 /*
391 * Check both floating and local interrupt's cr14 because
392 * bit IRQ_PEND_MCHK_REP could be set in both cases.
393 */
394 if (!(vcpu->arch.sie_block->gcr[14] &
395 (vcpu->kvm->arch.float_int.mchk.cr14 |
396 vcpu->arch.local_int.irq.mchk.cr14)))
397 __clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
398
399 /*
400 * STOP irqs will never be actively delivered. They are triggered via
401 * intercept requests and cleared when the stop intercept is performed.
402 */
403 __clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
404
405 return active_mask;
406}
407
408static void __set_cpu_idle(struct kvm_vcpu *vcpu)
409{
410 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
411 set_bit(vcpu->vcpu_id, vcpu->kvm->arch.idle_mask);
412}
413
414static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
415{
416 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
417 clear_bit(vcpu->vcpu_id, vcpu->kvm->arch.idle_mask);
418}
419
420static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
421{
422 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT |
423 CPUSTAT_STOP_INT);
424 vcpu->arch.sie_block->lctl = 0x0000;
425 vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
426
427 if (guestdbg_enabled(vcpu)) {
428 vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
429 LCTL_CR10 | LCTL_CR11);
430 vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
431 }
432}
433
434static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
435{
436 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK))
437 return;
438 if (psw_ioint_disabled(vcpu))
439 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT);
440 else
441 vcpu->arch.sie_block->lctl |= LCTL_CR6;
442}
443
444static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
445{
446 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK))
447 return;
448 if (psw_extint_disabled(vcpu))
449 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
450 else
451 vcpu->arch.sie_block->lctl |= LCTL_CR0;
452}
453
454static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
455{
456 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK))
457 return;
458 if (psw_mchk_disabled(vcpu))
459 vcpu->arch.sie_block->ictl |= ICTL_LPSW;
460 else
461 vcpu->arch.sie_block->lctl |= LCTL_CR14;
462}
463
464static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
465{
466 if (kvm_s390_is_stop_irq_pending(vcpu))
467 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
468}
469
470/* Set interception request for non-deliverable interrupts */
471static void set_intercept_indicators(struct kvm_vcpu *vcpu)
472{
473 set_intercept_indicators_io(vcpu);
474 set_intercept_indicators_ext(vcpu);
475 set_intercept_indicators_mchk(vcpu);
476 set_intercept_indicators_stop(vcpu);
477}
478
479static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
480{
481 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
482 int rc;
483
484 vcpu->stat.deliver_cputm++;
485 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
486 0, 0);
487
488 rc = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
489 (u16 *)__LC_EXT_INT_CODE);
490 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
491 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
492 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
493 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
494 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
495 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
496 return rc ? -EFAULT : 0;
497}
498
499static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
500{
501 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
502 int rc;
503
504 vcpu->stat.deliver_ckc++;
505 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
506 0, 0);
507
508 rc = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
509 (u16 __user *)__LC_EXT_INT_CODE);
510 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
511 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
512 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
513 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
514 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
515 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
516 return rc ? -EFAULT : 0;
517}
518
519static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
520{
521 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
522 struct kvm_s390_ext_info ext;
523 int rc;
524
525 spin_lock(&li->lock);
526 ext = li->irq.ext;
527 clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
528 li->irq.ext.ext_params2 = 0;
529 spin_unlock(&li->lock);
530
531 VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
532 ext.ext_params2);
533 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
534 KVM_S390_INT_PFAULT_INIT,
535 0, ext.ext_params2);
536
537 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
538 rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
539 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
540 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
541 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
542 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
543 rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
544 return rc ? -EFAULT : 0;
545}
546
547static int __write_machine_check(struct kvm_vcpu *vcpu,
548 struct kvm_s390_mchk_info *mchk)
549{
550 unsigned long ext_sa_addr;
551 unsigned long lc;
552 freg_t fprs[NUM_FPRS];
553 union mci mci;
554 int rc;
555
556 mci.val = mchk->mcic;
557 /* take care of lazy register loading */
558 save_fpu_regs();
559 save_access_regs(vcpu->run->s.regs.acrs);
560 if (MACHINE_HAS_GS && vcpu->arch.gs_enabled)
561 save_gs_cb(current->thread.gs_cb);
562
563 /* Extended save area */
564 rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
565 sizeof(unsigned long));
566 /* Only bits 0 through 63-LC are used for address formation */
567 lc = ext_sa_addr & MCESA_LC_MASK;
568 if (test_kvm_facility(vcpu->kvm, 133)) {
569 switch (lc) {
570 case 0:
571 case 10:
572 ext_sa_addr &= ~0x3ffUL;
573 break;
574 case 11:
575 ext_sa_addr &= ~0x7ffUL;
576 break;
577 case 12:
578 ext_sa_addr &= ~0xfffUL;
579 break;
580 default:
581 ext_sa_addr = 0;
582 break;
583 }
584 } else {
585 ext_sa_addr &= ~0x3ffUL;
586 }
587
588 if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
589 if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
590 512))
591 mci.vr = 0;
592 } else {
593 mci.vr = 0;
594 }
595 if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
596 && (lc == 11 || lc == 12)) {
597 if (write_guest_abs(vcpu, ext_sa_addr + 1024,
598 &vcpu->run->s.regs.gscb, 32))
599 mci.gs = 0;
600 } else {
601 mci.gs = 0;
602 }
603
604 /* General interruption information */
605 rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
606 rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
607 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
608 rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
609 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
610 rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
611
612 /* Register-save areas */
613 if (MACHINE_HAS_VX) {
614 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
615 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
616 } else {
617 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
618 vcpu->run->s.regs.fprs, 128);
619 }
620 rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
621 vcpu->run->s.regs.gprs, 128);
622 rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
623 (u32 __user *) __LC_FP_CREG_SAVE_AREA);
624 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
625 (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
626 rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
627 (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
628 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
629 (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
630 rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
631 &vcpu->run->s.regs.acrs, 64);
632 rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
633 &vcpu->arch.sie_block->gcr, 128);
634
635 /* Extended interruption information */
636 rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
637 (u32 __user *) __LC_EXT_DAMAGE_CODE);
638 rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
639 (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
640 rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
641 sizeof(mchk->fixed_logout));
642 return rc ? -EFAULT : 0;
643}
644
645static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
646{
647 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
648 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
649 struct kvm_s390_mchk_info mchk = {};
650 int deliver = 0;
651 int rc = 0;
652
653 spin_lock(&fi->lock);
654 spin_lock(&li->lock);
655 if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
656 test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
657 /*
658 * If there was an exigent machine check pending, then any
659 * repressible machine checks that might have been pending
660 * are indicated along with it, so always clear bits for
661 * repressible and exigent interrupts
662 */
663 mchk = li->irq.mchk;
664 clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
665 clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
666 memset(&li->irq.mchk, 0, sizeof(mchk));
667 deliver = 1;
668 }
669 /*
670 * We indicate floating repressible conditions along with
671 * other pending conditions. Channel Report Pending and Channel
672 * Subsystem damage are the only two and and are indicated by
673 * bits in mcic and masked in cr14.
674 */
675 if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
676 mchk.mcic |= fi->mchk.mcic;
677 mchk.cr14 |= fi->mchk.cr14;
678 memset(&fi->mchk, 0, sizeof(mchk));
679 deliver = 1;
680 }
681 spin_unlock(&li->lock);
682 spin_unlock(&fi->lock);
683
684 if (deliver) {
685 VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
686 mchk.mcic);
687 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
688 KVM_S390_MCHK,
689 mchk.cr14, mchk.mcic);
690 vcpu->stat.deliver_machine_check++;
691 rc = __write_machine_check(vcpu, &mchk);
692 }
693 return rc;
694}
695
696static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
697{
698 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
699 int rc;
700
701 VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
702 vcpu->stat.deliver_restart_signal++;
703 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
704
705 rc = write_guest_lc(vcpu,
706 offsetof(struct lowcore, restart_old_psw),
707 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
708 rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
709 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
710 clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
711 return rc ? -EFAULT : 0;
712}
713
714static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
715{
716 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
717 struct kvm_s390_prefix_info prefix;
718
719 spin_lock(&li->lock);
720 prefix = li->irq.prefix;
721 li->irq.prefix.address = 0;
722 clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
723 spin_unlock(&li->lock);
724
725 vcpu->stat.deliver_prefix_signal++;
726 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
727 KVM_S390_SIGP_SET_PREFIX,
728 prefix.address, 0);
729
730 kvm_s390_set_prefix(vcpu, prefix.address);
731 return 0;
732}
733
734static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
735{
736 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
737 int rc;
738 int cpu_addr;
739
740 spin_lock(&li->lock);
741 cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
742 clear_bit(cpu_addr, li->sigp_emerg_pending);
743 if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
744 clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
745 spin_unlock(&li->lock);
746
747 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
748 vcpu->stat.deliver_emergency_signal++;
749 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
750 cpu_addr, 0);
751
752 rc = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
753 (u16 *)__LC_EXT_INT_CODE);
754 rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
755 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
756 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
757 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
758 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
759 return rc ? -EFAULT : 0;
760}
761
762static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
763{
764 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
765 struct kvm_s390_extcall_info extcall;
766 int rc;
767
768 spin_lock(&li->lock);
769 extcall = li->irq.extcall;
770 li->irq.extcall.code = 0;
771 clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
772 spin_unlock(&li->lock);
773
774 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
775 vcpu->stat.deliver_external_call++;
776 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
777 KVM_S390_INT_EXTERNAL_CALL,
778 extcall.code, 0);
779
780 rc = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
781 (u16 *)__LC_EXT_INT_CODE);
782 rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
783 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
784 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
785 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
786 sizeof(psw_t));
787 return rc ? -EFAULT : 0;
788}
789
790static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
791{
792 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
793 struct kvm_s390_pgm_info pgm_info;
794 int rc = 0, nullifying = false;
795 u16 ilen;
796
797 spin_lock(&li->lock);
798 pgm_info = li->irq.pgm;
799 clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
800 memset(&li->irq.pgm, 0, sizeof(pgm_info));
801 spin_unlock(&li->lock);
802
803 ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
804 VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
805 pgm_info.code, ilen);
806 vcpu->stat.deliver_program++;
807 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
808 pgm_info.code, 0);
809
810 switch (pgm_info.code & ~PGM_PER) {
811 case PGM_AFX_TRANSLATION:
812 case PGM_ASX_TRANSLATION:
813 case PGM_EX_TRANSLATION:
814 case PGM_LFX_TRANSLATION:
815 case PGM_LSTE_SEQUENCE:
816 case PGM_LSX_TRANSLATION:
817 case PGM_LX_TRANSLATION:
818 case PGM_PRIMARY_AUTHORITY:
819 case PGM_SECONDARY_AUTHORITY:
820 nullifying = true;
821 /* fall through */
822 case PGM_SPACE_SWITCH:
823 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
824 (u64 *)__LC_TRANS_EXC_CODE);
825 break;
826 case PGM_ALEN_TRANSLATION:
827 case PGM_ALE_SEQUENCE:
828 case PGM_ASTE_INSTANCE:
829 case PGM_ASTE_SEQUENCE:
830 case PGM_ASTE_VALIDITY:
831 case PGM_EXTENDED_AUTHORITY:
832 rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
833 (u8 *)__LC_EXC_ACCESS_ID);
834 nullifying = true;
835 break;
836 case PGM_ASCE_TYPE:
837 case PGM_PAGE_TRANSLATION:
838 case PGM_REGION_FIRST_TRANS:
839 case PGM_REGION_SECOND_TRANS:
840 case PGM_REGION_THIRD_TRANS:
841 case PGM_SEGMENT_TRANSLATION:
842 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
843 (u64 *)__LC_TRANS_EXC_CODE);
844 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
845 (u8 *)__LC_EXC_ACCESS_ID);
846 rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
847 (u8 *)__LC_OP_ACCESS_ID);
848 nullifying = true;
849 break;
850 case PGM_MONITOR:
851 rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
852 (u16 *)__LC_MON_CLASS_NR);
853 rc |= put_guest_lc(vcpu, pgm_info.mon_code,
854 (u64 *)__LC_MON_CODE);
855 break;
856 case PGM_VECTOR_PROCESSING:
857 case PGM_DATA:
858 rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
859 (u32 *)__LC_DATA_EXC_CODE);
860 break;
861 case PGM_PROTECTION:
862 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
863 (u64 *)__LC_TRANS_EXC_CODE);
864 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
865 (u8 *)__LC_EXC_ACCESS_ID);
866 break;
867 case PGM_STACK_FULL:
868 case PGM_STACK_EMPTY:
869 case PGM_STACK_SPECIFICATION:
870 case PGM_STACK_TYPE:
871 case PGM_STACK_OPERATION:
872 case PGM_TRACE_TABEL:
873 case PGM_CRYPTO_OPERATION:
874 nullifying = true;
875 break;
876 }
877
878 if (pgm_info.code & PGM_PER) {
879 rc |= put_guest_lc(vcpu, pgm_info.per_code,
880 (u8 *) __LC_PER_CODE);
881 rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
882 (u8 *)__LC_PER_ATMID);
883 rc |= put_guest_lc(vcpu, pgm_info.per_address,
884 (u64 *) __LC_PER_ADDRESS);
885 rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
886 (u8 *) __LC_PER_ACCESS_ID);
887 }
888
889 if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
890 kvm_s390_rewind_psw(vcpu, ilen);
891
892 /* bit 1+2 of the target are the ilc, so we can directly use ilen */
893 rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
894 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
895 (u64 *) __LC_LAST_BREAK);
896 rc |= put_guest_lc(vcpu, pgm_info.code,
897 (u16 *)__LC_PGM_INT_CODE);
898 rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
899 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
900 rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
901 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
902 return rc ? -EFAULT : 0;
903}
904
905static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
906{
907 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
908 struct kvm_s390_ext_info ext;
909 int rc = 0;
910
911 spin_lock(&fi->lock);
912 if (!(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
913 spin_unlock(&fi->lock);
914 return 0;
915 }
916 ext = fi->srv_signal;
917 memset(&fi->srv_signal, 0, sizeof(ext));
918 clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
919 spin_unlock(&fi->lock);
920
921 VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
922 ext.ext_params);
923 vcpu->stat.deliver_service_signal++;
924 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
925 ext.ext_params, 0);
926
927 rc = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
928 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
929 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
930 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
931 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
932 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
933 rc |= put_guest_lc(vcpu, ext.ext_params,
934 (u32 *)__LC_EXT_PARAMS);
935
936 return rc ? -EFAULT : 0;
937}
938
939static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
940{
941 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
942 struct kvm_s390_interrupt_info *inti;
943 int rc = 0;
944
945 spin_lock(&fi->lock);
946 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
947 struct kvm_s390_interrupt_info,
948 list);
949 if (inti) {
950 list_del(&inti->list);
951 fi->counters[FIRQ_CNTR_PFAULT] -= 1;
952 }
953 if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
954 clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
955 spin_unlock(&fi->lock);
956
957 if (inti) {
958 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
959 KVM_S390_INT_PFAULT_DONE, 0,
960 inti->ext.ext_params2);
961 VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
962 inti->ext.ext_params2);
963
964 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
965 (u16 *)__LC_EXT_INT_CODE);
966 rc |= put_guest_lc(vcpu, PFAULT_DONE,
967 (u16 *)__LC_EXT_CPU_ADDR);
968 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
969 &vcpu->arch.sie_block->gpsw,
970 sizeof(psw_t));
971 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
972 &vcpu->arch.sie_block->gpsw,
973 sizeof(psw_t));
974 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
975 (u64 *)__LC_EXT_PARAMS2);
976 kfree(inti);
977 }
978 return rc ? -EFAULT : 0;
979}
980
981static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
982{
983 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
984 struct kvm_s390_interrupt_info *inti;
985 int rc = 0;
986
987 spin_lock(&fi->lock);
988 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
989 struct kvm_s390_interrupt_info,
990 list);
991 if (inti) {
992 VCPU_EVENT(vcpu, 4,
993 "deliver: virtio parm: 0x%x,parm64: 0x%llx",
994 inti->ext.ext_params, inti->ext.ext_params2);
995 vcpu->stat.deliver_virtio++;
996 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
997 inti->type,
998 inti->ext.ext_params,
999 inti->ext.ext_params2);
1000 list_del(&inti->list);
1001 fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
1002 }
1003 if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
1004 clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1005 spin_unlock(&fi->lock);
1006
1007 if (inti) {
1008 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1009 (u16 *)__LC_EXT_INT_CODE);
1010 rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
1011 (u16 *)__LC_EXT_CPU_ADDR);
1012 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1013 &vcpu->arch.sie_block->gpsw,
1014 sizeof(psw_t));
1015 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1016 &vcpu->arch.sie_block->gpsw,
1017 sizeof(psw_t));
1018 rc |= put_guest_lc(vcpu, inti->ext.ext_params,
1019 (u32 *)__LC_EXT_PARAMS);
1020 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1021 (u64 *)__LC_EXT_PARAMS2);
1022 kfree(inti);
1023 }
1024 return rc ? -EFAULT : 0;
1025}
1026
1027static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io)
1028{
1029 int rc;
1030
1031 rc = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID);
1032 rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR);
1033 rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM);
1034 rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD);
1035 rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
1036 &vcpu->arch.sie_block->gpsw,
1037 sizeof(psw_t));
1038 rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
1039 &vcpu->arch.sie_block->gpsw,
1040 sizeof(psw_t));
1041 return rc ? -EFAULT : 0;
1042}
1043
1044static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
1045 unsigned long irq_type)
1046{
1047 struct list_head *isc_list;
1048 struct kvm_s390_float_interrupt *fi;
1049 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1050 struct kvm_s390_interrupt_info *inti = NULL;
1051 struct kvm_s390_io_info io;
1052 u32 isc;
1053 int rc = 0;
1054
1055 fi = &vcpu->kvm->arch.float_int;
1056
1057 spin_lock(&fi->lock);
1058 isc = irq_type_to_isc(irq_type);
1059 isc_list = &fi->lists[isc];
1060 inti = list_first_entry_or_null(isc_list,
1061 struct kvm_s390_interrupt_info,
1062 list);
1063 if (inti) {
1064 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1065 VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
1066 else
1067 VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
1068 inti->io.subchannel_id >> 8,
1069 inti->io.subchannel_id >> 1 & 0x3,
1070 inti->io.subchannel_nr);
1071
1072 vcpu->stat.deliver_io++;
1073 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1074 inti->type,
1075 ((__u32)inti->io.subchannel_id << 16) |
1076 inti->io.subchannel_nr,
1077 ((__u64)inti->io.io_int_parm << 32) |
1078 inti->io.io_int_word);
1079 list_del(&inti->list);
1080 fi->counters[FIRQ_CNTR_IO] -= 1;
1081 }
1082 if (list_empty(isc_list))
1083 clear_bit(irq_type, &fi->pending_irqs);
1084 spin_unlock(&fi->lock);
1085
1086 if (inti) {
1087 rc = __do_deliver_io(vcpu, &(inti->io));
1088 kfree(inti);
1089 goto out;
1090 }
1091
1092 if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) {
1093 /*
1094 * in case an adapter interrupt was not delivered
1095 * in SIE context KVM will handle the delivery
1096 */
1097 VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc);
1098 memset(&io, 0, sizeof(io));
1099 io.io_int_word = isc_to_int_word(isc);
1100 vcpu->stat.deliver_io++;
1101 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1102 KVM_S390_INT_IO(1, 0, 0, 0),
1103 ((__u32)io.subchannel_id << 16) |
1104 io.subchannel_nr,
1105 ((__u64)io.io_int_parm << 32) |
1106 io.io_int_word);
1107 rc = __do_deliver_io(vcpu, &io);
1108 }
1109out:
1110 return rc;
1111}
1112
1113/* Check whether an external call is pending (deliverable or not) */
1114int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
1115{
1116 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1117
1118 if (!sclp.has_sigpif)
1119 return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
1120
1121 return sca_ext_call_pending(vcpu, NULL);
1122}
1123
1124int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
1125{
1126 if (deliverable_irqs(vcpu))
1127 return 1;
1128
1129 if (kvm_cpu_has_pending_timer(vcpu))
1130 return 1;
1131
1132 /* external call pending and deliverable */
1133 if (kvm_s390_ext_call_pending(vcpu) &&
1134 !psw_extint_disabled(vcpu) &&
1135 (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
1136 return 1;
1137
1138 if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1139 return 1;
1140 return 0;
1141}
1142
1143int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1144{
1145 return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1146}
1147
1148static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1149{
1150 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1151 const u64 ckc = vcpu->arch.sie_block->ckc;
1152 u64 cputm, sltime = 0;
1153
1154 if (ckc_interrupts_enabled(vcpu)) {
1155 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
1156 if ((s64)now < (s64)ckc)
1157 sltime = tod_to_ns((s64)ckc - (s64)now);
1158 } else if (now < ckc) {
1159 sltime = tod_to_ns(ckc - now);
1160 }
1161 /* already expired */
1162 if (!sltime)
1163 return 0;
1164 if (cpu_timer_interrupts_enabled(vcpu)) {
1165 cputm = kvm_s390_get_cpu_timer(vcpu);
1166 /* already expired? */
1167 if (cputm >> 63)
1168 return 0;
1169 return min(sltime, tod_to_ns(cputm));
1170 }
1171 } else if (cpu_timer_interrupts_enabled(vcpu)) {
1172 sltime = kvm_s390_get_cpu_timer(vcpu);
1173 /* already expired? */
1174 if (sltime >> 63)
1175 return 0;
1176 }
1177 return sltime;
1178}
1179
1180int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1181{
1182 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1183 u64 sltime;
1184
1185 vcpu->stat.exit_wait_state++;
1186
1187 /* fast path */
1188 if (kvm_arch_vcpu_runnable(vcpu))
1189 return 0;
1190
1191 if (psw_interrupts_disabled(vcpu)) {
1192 VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1193 return -EOPNOTSUPP; /* disabled wait */
1194 }
1195
1196 if (gi->origin &&
1197 (gisa_get_ipm_or_restore_iam(gi) &
1198 vcpu->arch.sie_block->gcr[6] >> 24))
1199 return 0;
1200
1201 if (!ckc_interrupts_enabled(vcpu) &&
1202 !cpu_timer_interrupts_enabled(vcpu)) {
1203 VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1204 __set_cpu_idle(vcpu);
1205 goto no_timer;
1206 }
1207
1208 sltime = __calculate_sltime(vcpu);
1209 if (!sltime)
1210 return 0;
1211
1212 __set_cpu_idle(vcpu);
1213 hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1214 VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1215no_timer:
1216 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1217 kvm_vcpu_block(vcpu);
1218 __unset_cpu_idle(vcpu);
1219 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1220
1221 hrtimer_cancel(&vcpu->arch.ckc_timer);
1222 return 0;
1223}
1224
1225void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1226{
1227 vcpu->valid_wakeup = true;
1228 kvm_vcpu_wake_up(vcpu);
1229
1230 /*
1231 * The VCPU might not be sleeping but rather executing VSIE. Let's
1232 * kick it, so it leaves the SIE to process the request.
1233 */
1234 kvm_s390_vsie_kick(vcpu);
1235}
1236
1237enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1238{
1239 struct kvm_vcpu *vcpu;
1240 u64 sltime;
1241
1242 vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1243 sltime = __calculate_sltime(vcpu);
1244
1245 /*
1246 * If the monotonic clock runs faster than the tod clock we might be
1247 * woken up too early and have to go back to sleep to avoid deadlocks.
1248 */
1249 if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1250 return HRTIMER_RESTART;
1251 kvm_s390_vcpu_wakeup(vcpu);
1252 return HRTIMER_NORESTART;
1253}
1254
1255void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1256{
1257 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1258
1259 spin_lock(&li->lock);
1260 li->pending_irqs = 0;
1261 bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1262 memset(&li->irq, 0, sizeof(li->irq));
1263 spin_unlock(&li->lock);
1264
1265 sca_clear_ext_call(vcpu);
1266}
1267
1268int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1269{
1270 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1271 int rc = 0;
1272 unsigned long irq_type;
1273 unsigned long irqs;
1274
1275 __reset_intercept_indicators(vcpu);
1276
1277 /* pending ckc conditions might have been invalidated */
1278 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1279 if (ckc_irq_pending(vcpu))
1280 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1281
1282 /* pending cpu timer conditions might have been invalidated */
1283 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1284 if (cpu_timer_irq_pending(vcpu))
1285 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1286
1287 while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1288 /* bits are in the reverse order of interrupt priority */
1289 irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1290 switch (irq_type) {
1291 case IRQ_PEND_IO_ISC_0:
1292 case IRQ_PEND_IO_ISC_1:
1293 case IRQ_PEND_IO_ISC_2:
1294 case IRQ_PEND_IO_ISC_3:
1295 case IRQ_PEND_IO_ISC_4:
1296 case IRQ_PEND_IO_ISC_5:
1297 case IRQ_PEND_IO_ISC_6:
1298 case IRQ_PEND_IO_ISC_7:
1299 rc = __deliver_io(vcpu, irq_type);
1300 break;
1301 case IRQ_PEND_MCHK_EX:
1302 case IRQ_PEND_MCHK_REP:
1303 rc = __deliver_machine_check(vcpu);
1304 break;
1305 case IRQ_PEND_PROG:
1306 rc = __deliver_prog(vcpu);
1307 break;
1308 case IRQ_PEND_EXT_EMERGENCY:
1309 rc = __deliver_emergency_signal(vcpu);
1310 break;
1311 case IRQ_PEND_EXT_EXTERNAL:
1312 rc = __deliver_external_call(vcpu);
1313 break;
1314 case IRQ_PEND_EXT_CLOCK_COMP:
1315 rc = __deliver_ckc(vcpu);
1316 break;
1317 case IRQ_PEND_EXT_CPU_TIMER:
1318 rc = __deliver_cpu_timer(vcpu);
1319 break;
1320 case IRQ_PEND_RESTART:
1321 rc = __deliver_restart(vcpu);
1322 break;
1323 case IRQ_PEND_SET_PREFIX:
1324 rc = __deliver_set_prefix(vcpu);
1325 break;
1326 case IRQ_PEND_PFAULT_INIT:
1327 rc = __deliver_pfault_init(vcpu);
1328 break;
1329 case IRQ_PEND_EXT_SERVICE:
1330 rc = __deliver_service(vcpu);
1331 break;
1332 case IRQ_PEND_PFAULT_DONE:
1333 rc = __deliver_pfault_done(vcpu);
1334 break;
1335 case IRQ_PEND_VIRTIO:
1336 rc = __deliver_virtio(vcpu);
1337 break;
1338 default:
1339 WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1340 clear_bit(irq_type, &li->pending_irqs);
1341 }
1342 }
1343
1344 set_intercept_indicators(vcpu);
1345
1346 return rc;
1347}
1348
1349static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1350{
1351 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1352
1353 vcpu->stat.inject_program++;
1354 VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1355 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1356 irq->u.pgm.code, 0);
1357
1358 if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1359 /* auto detection if no valid ILC was given */
1360 irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1361 irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1362 irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1363 }
1364
1365 if (irq->u.pgm.code == PGM_PER) {
1366 li->irq.pgm.code |= PGM_PER;
1367 li->irq.pgm.flags = irq->u.pgm.flags;
1368 /* only modify PER related information */
1369 li->irq.pgm.per_address = irq->u.pgm.per_address;
1370 li->irq.pgm.per_code = irq->u.pgm.per_code;
1371 li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1372 li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1373 } else if (!(irq->u.pgm.code & PGM_PER)) {
1374 li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1375 irq->u.pgm.code;
1376 li->irq.pgm.flags = irq->u.pgm.flags;
1377 /* only modify non-PER information */
1378 li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1379 li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1380 li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1381 li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1382 li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1383 li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1384 } else {
1385 li->irq.pgm = irq->u.pgm;
1386 }
1387 set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1388 return 0;
1389}
1390
1391static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1392{
1393 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1394
1395 vcpu->stat.inject_pfault_init++;
1396 VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1397 irq->u.ext.ext_params2);
1398 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1399 irq->u.ext.ext_params,
1400 irq->u.ext.ext_params2);
1401
1402 li->irq.ext = irq->u.ext;
1403 set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1404 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1405 return 0;
1406}
1407
1408static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1409{
1410 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1411 struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1412 uint16_t src_id = irq->u.extcall.code;
1413
1414 vcpu->stat.inject_external_call++;
1415 VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1416 src_id);
1417 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1418 src_id, 0);
1419
1420 /* sending vcpu invalid */
1421 if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1422 return -EINVAL;
1423
1424 if (sclp.has_sigpif)
1425 return sca_inject_ext_call(vcpu, src_id);
1426
1427 if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1428 return -EBUSY;
1429 *extcall = irq->u.extcall;
1430 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1431 return 0;
1432}
1433
1434static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1435{
1436 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1437 struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1438
1439 vcpu->stat.inject_set_prefix++;
1440 VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1441 irq->u.prefix.address);
1442 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1443 irq->u.prefix.address, 0);
1444
1445 if (!is_vcpu_stopped(vcpu))
1446 return -EBUSY;
1447
1448 *prefix = irq->u.prefix;
1449 set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1450 return 0;
1451}
1452
1453#define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1454static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1455{
1456 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1457 struct kvm_s390_stop_info *stop = &li->irq.stop;
1458 int rc = 0;
1459
1460 vcpu->stat.inject_stop_signal++;
1461 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1462
1463 if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1464 return -EINVAL;
1465
1466 if (is_vcpu_stopped(vcpu)) {
1467 if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1468 rc = kvm_s390_store_status_unloaded(vcpu,
1469 KVM_S390_STORE_STATUS_NOADDR);
1470 return rc;
1471 }
1472
1473 if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1474 return -EBUSY;
1475 stop->flags = irq->u.stop.flags;
1476 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
1477 return 0;
1478}
1479
1480static int __inject_sigp_restart(struct kvm_vcpu *vcpu,
1481 struct kvm_s390_irq *irq)
1482{
1483 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1484
1485 vcpu->stat.inject_restart++;
1486 VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1487 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1488
1489 set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1490 return 0;
1491}
1492
1493static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1494 struct kvm_s390_irq *irq)
1495{
1496 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1497
1498 vcpu->stat.inject_emergency_signal++;
1499 VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1500 irq->u.emerg.code);
1501 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1502 irq->u.emerg.code, 0);
1503
1504 /* sending vcpu invalid */
1505 if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1506 return -EINVAL;
1507
1508 set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1509 set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1510 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1511 return 0;
1512}
1513
1514static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1515{
1516 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1517 struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1518
1519 vcpu->stat.inject_mchk++;
1520 VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1521 irq->u.mchk.mcic);
1522 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1523 irq->u.mchk.mcic);
1524
1525 /*
1526 * Because repressible machine checks can be indicated along with
1527 * exigent machine checks (PoP, Chapter 11, Interruption action)
1528 * we need to combine cr14, mcic and external damage code.
1529 * Failing storage address and the logout area should not be or'ed
1530 * together, we just indicate the last occurrence of the corresponding
1531 * machine check
1532 */
1533 mchk->cr14 |= irq->u.mchk.cr14;
1534 mchk->mcic |= irq->u.mchk.mcic;
1535 mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1536 mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1537 memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1538 sizeof(mchk->fixed_logout));
1539 if (mchk->mcic & MCHK_EX_MASK)
1540 set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1541 else if (mchk->mcic & MCHK_REP_MASK)
1542 set_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
1543 return 0;
1544}
1545
1546static int __inject_ckc(struct kvm_vcpu *vcpu)
1547{
1548 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1549
1550 vcpu->stat.inject_ckc++;
1551 VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1552 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1553 0, 0);
1554
1555 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1556 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1557 return 0;
1558}
1559
1560static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1561{
1562 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1563
1564 vcpu->stat.inject_cputm++;
1565 VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1566 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1567 0, 0);
1568
1569 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1570 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1571 return 0;
1572}
1573
1574static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1575 int isc, u32 schid)
1576{
1577 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1578 struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1579 struct kvm_s390_interrupt_info *iter;
1580 u16 id = (schid & 0xffff0000U) >> 16;
1581 u16 nr = schid & 0x0000ffffU;
1582
1583 spin_lock(&fi->lock);
1584 list_for_each_entry(iter, isc_list, list) {
1585 if (schid && (id != iter->io.subchannel_id ||
1586 nr != iter->io.subchannel_nr))
1587 continue;
1588 /* found an appropriate entry */
1589 list_del_init(&iter->list);
1590 fi->counters[FIRQ_CNTR_IO] -= 1;
1591 if (list_empty(isc_list))
1592 clear_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1593 spin_unlock(&fi->lock);
1594 return iter;
1595 }
1596 spin_unlock(&fi->lock);
1597 return NULL;
1598}
1599
1600static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm,
1601 u64 isc_mask, u32 schid)
1602{
1603 struct kvm_s390_interrupt_info *inti = NULL;
1604 int isc;
1605
1606 for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1607 if (isc_mask & isc_to_isc_bits(isc))
1608 inti = get_io_int(kvm, isc, schid);
1609 }
1610 return inti;
1611}
1612
1613static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid)
1614{
1615 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1616 unsigned long active_mask;
1617 int isc;
1618
1619 if (schid)
1620 goto out;
1621 if (!gi->origin)
1622 goto out;
1623
1624 active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32;
1625 while (active_mask) {
1626 isc = __fls(active_mask) ^ (BITS_PER_LONG - 1);
1627 if (gisa_tac_ipm_gisc(gi->origin, isc))
1628 return isc;
1629 clear_bit_inv(isc, &active_mask);
1630 }
1631out:
1632 return -EINVAL;
1633}
1634
1635/*
1636 * Dequeue and return an I/O interrupt matching any of the interruption
1637 * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1638 * Take into account the interrupts pending in the interrupt list and in GISA.
1639 *
1640 * Note that for a guest that does not enable I/O interrupts
1641 * but relies on TPI, a flood of classic interrupts may starve
1642 * out adapter interrupts on the same isc. Linux does not do
1643 * that, and it is possible to work around the issue by configuring
1644 * different iscs for classic and adapter interrupts in the guest,
1645 * but we may want to revisit this in the future.
1646 */
1647struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1648 u64 isc_mask, u32 schid)
1649{
1650 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1651 struct kvm_s390_interrupt_info *inti, *tmp_inti;
1652 int isc;
1653
1654 inti = get_top_io_int(kvm, isc_mask, schid);
1655
1656 isc = get_top_gisa_isc(kvm, isc_mask, schid);
1657 if (isc < 0)
1658 /* no AI in GISA */
1659 goto out;
1660
1661 if (!inti)
1662 /* AI in GISA but no classical IO int */
1663 goto gisa_out;
1664
1665 /* both types of interrupts present */
1666 if (int_word_to_isc(inti->io.io_int_word) <= isc) {
1667 /* classical IO int with higher priority */
1668 gisa_set_ipm_gisc(gi->origin, isc);
1669 goto out;
1670 }
1671gisa_out:
1672 tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1673 if (tmp_inti) {
1674 tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0);
1675 tmp_inti->io.io_int_word = isc_to_int_word(isc);
1676 if (inti)
1677 kvm_s390_reinject_io_int(kvm, inti);
1678 inti = tmp_inti;
1679 } else
1680 gisa_set_ipm_gisc(gi->origin, isc);
1681out:
1682 return inti;
1683}
1684
1685#define SCCB_MASK 0xFFFFFFF8
1686#define SCCB_EVENT_PENDING 0x3
1687
1688static int __inject_service(struct kvm *kvm,
1689 struct kvm_s390_interrupt_info *inti)
1690{
1691 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1692
1693 kvm->stat.inject_service_signal++;
1694 spin_lock(&fi->lock);
1695 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1696 /*
1697 * Early versions of the QEMU s390 bios will inject several
1698 * service interrupts after another without handling a
1699 * condition code indicating busy.
1700 * We will silently ignore those superfluous sccb values.
1701 * A future version of QEMU will take care of serialization
1702 * of servc requests
1703 */
1704 if (fi->srv_signal.ext_params & SCCB_MASK)
1705 goto out;
1706 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1707 set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1708out:
1709 spin_unlock(&fi->lock);
1710 kfree(inti);
1711 return 0;
1712}
1713
1714static int __inject_virtio(struct kvm *kvm,
1715 struct kvm_s390_interrupt_info *inti)
1716{
1717 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1718
1719 kvm->stat.inject_virtio++;
1720 spin_lock(&fi->lock);
1721 if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1722 spin_unlock(&fi->lock);
1723 return -EBUSY;
1724 }
1725 fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1726 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1727 set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1728 spin_unlock(&fi->lock);
1729 return 0;
1730}
1731
1732static int __inject_pfault_done(struct kvm *kvm,
1733 struct kvm_s390_interrupt_info *inti)
1734{
1735 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1736
1737 kvm->stat.inject_pfault_done++;
1738 spin_lock(&fi->lock);
1739 if (fi->counters[FIRQ_CNTR_PFAULT] >=
1740 (ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1741 spin_unlock(&fi->lock);
1742 return -EBUSY;
1743 }
1744 fi->counters[FIRQ_CNTR_PFAULT] += 1;
1745 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1746 set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1747 spin_unlock(&fi->lock);
1748 return 0;
1749}
1750
1751#define CR_PENDING_SUBCLASS 28
1752static int __inject_float_mchk(struct kvm *kvm,
1753 struct kvm_s390_interrupt_info *inti)
1754{
1755 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1756
1757 kvm->stat.inject_float_mchk++;
1758 spin_lock(&fi->lock);
1759 fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1760 fi->mchk.mcic |= inti->mchk.mcic;
1761 set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1762 spin_unlock(&fi->lock);
1763 kfree(inti);
1764 return 0;
1765}
1766
1767static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1768{
1769 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1770 struct kvm_s390_float_interrupt *fi;
1771 struct list_head *list;
1772 int isc;
1773
1774 kvm->stat.inject_io++;
1775 isc = int_word_to_isc(inti->io.io_int_word);
1776
1777 if (gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) {
1778 VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc);
1779 gisa_set_ipm_gisc(gi->origin, isc);
1780 kfree(inti);
1781 return 0;
1782 }
1783
1784 fi = &kvm->arch.float_int;
1785 spin_lock(&fi->lock);
1786 if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1787 spin_unlock(&fi->lock);
1788 return -EBUSY;
1789 }
1790 fi->counters[FIRQ_CNTR_IO] += 1;
1791
1792 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1793 VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1794 else
1795 VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1796 inti->io.subchannel_id >> 8,
1797 inti->io.subchannel_id >> 1 & 0x3,
1798 inti->io.subchannel_nr);
1799 list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1800 list_add_tail(&inti->list, list);
1801 set_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1802 spin_unlock(&fi->lock);
1803 return 0;
1804}
1805
1806/*
1807 * Find a destination VCPU for a floating irq and kick it.
1808 */
1809static void __floating_irq_kick(struct kvm *kvm, u64 type)
1810{
1811 struct kvm_vcpu *dst_vcpu;
1812 int sigcpu, online_vcpus, nr_tries = 0;
1813
1814 online_vcpus = atomic_read(&kvm->online_vcpus);
1815 if (!online_vcpus)
1816 return;
1817
1818 /* find idle VCPUs first, then round robin */
1819 sigcpu = find_first_bit(kvm->arch.idle_mask, online_vcpus);
1820 if (sigcpu == online_vcpus) {
1821 do {
1822 sigcpu = kvm->arch.float_int.next_rr_cpu++;
1823 kvm->arch.float_int.next_rr_cpu %= online_vcpus;
1824 /* avoid endless loops if all vcpus are stopped */
1825 if (nr_tries++ >= online_vcpus)
1826 return;
1827 } while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1828 }
1829 dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1830
1831 /* make the VCPU drop out of the SIE, or wake it up if sleeping */
1832 switch (type) {
1833 case KVM_S390_MCHK:
1834 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT);
1835 break;
1836 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1837 if (!(type & KVM_S390_INT_IO_AI_MASK &&
1838 kvm->arch.gisa_int.origin))
1839 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT);
1840 break;
1841 default:
1842 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT);
1843 break;
1844 }
1845 kvm_s390_vcpu_wakeup(dst_vcpu);
1846}
1847
1848static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1849{
1850 u64 type = READ_ONCE(inti->type);
1851 int rc;
1852
1853 switch (type) {
1854 case KVM_S390_MCHK:
1855 rc = __inject_float_mchk(kvm, inti);
1856 break;
1857 case KVM_S390_INT_VIRTIO:
1858 rc = __inject_virtio(kvm, inti);
1859 break;
1860 case KVM_S390_INT_SERVICE:
1861 rc = __inject_service(kvm, inti);
1862 break;
1863 case KVM_S390_INT_PFAULT_DONE:
1864 rc = __inject_pfault_done(kvm, inti);
1865 break;
1866 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1867 rc = __inject_io(kvm, inti);
1868 break;
1869 default:
1870 rc = -EINVAL;
1871 }
1872 if (rc)
1873 return rc;
1874
1875 __floating_irq_kick(kvm, type);
1876 return 0;
1877}
1878
1879int kvm_s390_inject_vm(struct kvm *kvm,
1880 struct kvm_s390_interrupt *s390int)
1881{
1882 struct kvm_s390_interrupt_info *inti;
1883 int rc;
1884
1885 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1886 if (!inti)
1887 return -ENOMEM;
1888
1889 inti->type = s390int->type;
1890 switch (inti->type) {
1891 case KVM_S390_INT_VIRTIO:
1892 VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
1893 s390int->parm, s390int->parm64);
1894 inti->ext.ext_params = s390int->parm;
1895 inti->ext.ext_params2 = s390int->parm64;
1896 break;
1897 case KVM_S390_INT_SERVICE:
1898 VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
1899 inti->ext.ext_params = s390int->parm;
1900 break;
1901 case KVM_S390_INT_PFAULT_DONE:
1902 inti->ext.ext_params2 = s390int->parm64;
1903 break;
1904 case KVM_S390_MCHK:
1905 VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
1906 s390int->parm64);
1907 inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
1908 inti->mchk.mcic = s390int->parm64;
1909 break;
1910 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1911 inti->io.subchannel_id = s390int->parm >> 16;
1912 inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
1913 inti->io.io_int_parm = s390int->parm64 >> 32;
1914 inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
1915 break;
1916 default:
1917 kfree(inti);
1918 return -EINVAL;
1919 }
1920 trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
1921 2);
1922
1923 rc = __inject_vm(kvm, inti);
1924 if (rc)
1925 kfree(inti);
1926 return rc;
1927}
1928
1929int kvm_s390_reinject_io_int(struct kvm *kvm,
1930 struct kvm_s390_interrupt_info *inti)
1931{
1932 return __inject_vm(kvm, inti);
1933}
1934
1935int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
1936 struct kvm_s390_irq *irq)
1937{
1938 irq->type = s390int->type;
1939 switch (irq->type) {
1940 case KVM_S390_PROGRAM_INT:
1941 if (s390int->parm & 0xffff0000)
1942 return -EINVAL;
1943 irq->u.pgm.code = s390int->parm;
1944 break;
1945 case KVM_S390_SIGP_SET_PREFIX:
1946 irq->u.prefix.address = s390int->parm;
1947 break;
1948 case KVM_S390_SIGP_STOP:
1949 irq->u.stop.flags = s390int->parm;
1950 break;
1951 case KVM_S390_INT_EXTERNAL_CALL:
1952 if (s390int->parm & 0xffff0000)
1953 return -EINVAL;
1954 irq->u.extcall.code = s390int->parm;
1955 break;
1956 case KVM_S390_INT_EMERGENCY:
1957 if (s390int->parm & 0xffff0000)
1958 return -EINVAL;
1959 irq->u.emerg.code = s390int->parm;
1960 break;
1961 case KVM_S390_MCHK:
1962 irq->u.mchk.mcic = s390int->parm64;
1963 break;
1964 case KVM_S390_INT_PFAULT_INIT:
1965 irq->u.ext.ext_params = s390int->parm;
1966 irq->u.ext.ext_params2 = s390int->parm64;
1967 break;
1968 case KVM_S390_RESTART:
1969 case KVM_S390_INT_CLOCK_COMP:
1970 case KVM_S390_INT_CPU_TIMER:
1971 break;
1972 default:
1973 return -EINVAL;
1974 }
1975 return 0;
1976}
1977
1978int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
1979{
1980 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1981
1982 return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1983}
1984
1985void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
1986{
1987 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1988
1989 spin_lock(&li->lock);
1990 li->irq.stop.flags = 0;
1991 clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1992 spin_unlock(&li->lock);
1993}
1994
1995static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1996{
1997 int rc;
1998
1999 switch (irq->type) {
2000 case KVM_S390_PROGRAM_INT:
2001 rc = __inject_prog(vcpu, irq);
2002 break;
2003 case KVM_S390_SIGP_SET_PREFIX:
2004 rc = __inject_set_prefix(vcpu, irq);
2005 break;
2006 case KVM_S390_SIGP_STOP:
2007 rc = __inject_sigp_stop(vcpu, irq);
2008 break;
2009 case KVM_S390_RESTART:
2010 rc = __inject_sigp_restart(vcpu, irq);
2011 break;
2012 case KVM_S390_INT_CLOCK_COMP:
2013 rc = __inject_ckc(vcpu);
2014 break;
2015 case KVM_S390_INT_CPU_TIMER:
2016 rc = __inject_cpu_timer(vcpu);
2017 break;
2018 case KVM_S390_INT_EXTERNAL_CALL:
2019 rc = __inject_extcall(vcpu, irq);
2020 break;
2021 case KVM_S390_INT_EMERGENCY:
2022 rc = __inject_sigp_emergency(vcpu, irq);
2023 break;
2024 case KVM_S390_MCHK:
2025 rc = __inject_mchk(vcpu, irq);
2026 break;
2027 case KVM_S390_INT_PFAULT_INIT:
2028 rc = __inject_pfault_init(vcpu, irq);
2029 break;
2030 case KVM_S390_INT_VIRTIO:
2031 case KVM_S390_INT_SERVICE:
2032 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2033 default:
2034 rc = -EINVAL;
2035 }
2036
2037 return rc;
2038}
2039
2040int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2041{
2042 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2043 int rc;
2044
2045 spin_lock(&li->lock);
2046 rc = do_inject_vcpu(vcpu, irq);
2047 spin_unlock(&li->lock);
2048 if (!rc)
2049 kvm_s390_vcpu_wakeup(vcpu);
2050 return rc;
2051}
2052
2053static inline void clear_irq_list(struct list_head *_list)
2054{
2055 struct kvm_s390_interrupt_info *inti, *n;
2056
2057 list_for_each_entry_safe(inti, n, _list, list) {
2058 list_del(&inti->list);
2059 kfree(inti);
2060 }
2061}
2062
2063static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
2064 struct kvm_s390_irq *irq)
2065{
2066 irq->type = inti->type;
2067 switch (inti->type) {
2068 case KVM_S390_INT_PFAULT_INIT:
2069 case KVM_S390_INT_PFAULT_DONE:
2070 case KVM_S390_INT_VIRTIO:
2071 irq->u.ext = inti->ext;
2072 break;
2073 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2074 irq->u.io = inti->io;
2075 break;
2076 }
2077}
2078
2079void kvm_s390_clear_float_irqs(struct kvm *kvm)
2080{
2081 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2082 int i;
2083
2084 spin_lock(&fi->lock);
2085 fi->pending_irqs = 0;
2086 memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
2087 memset(&fi->mchk, 0, sizeof(fi->mchk));
2088 for (i = 0; i < FIRQ_LIST_COUNT; i++)
2089 clear_irq_list(&fi->lists[i]);
2090 for (i = 0; i < FIRQ_MAX_COUNT; i++)
2091 fi->counters[i] = 0;
2092 spin_unlock(&fi->lock);
2093 kvm_s390_gisa_clear(kvm);
2094};
2095
2096static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
2097{
2098 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2099 struct kvm_s390_interrupt_info *inti;
2100 struct kvm_s390_float_interrupt *fi;
2101 struct kvm_s390_irq *buf;
2102 struct kvm_s390_irq *irq;
2103 int max_irqs;
2104 int ret = 0;
2105 int n = 0;
2106 int i;
2107
2108 if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
2109 return -EINVAL;
2110
2111 /*
2112 * We are already using -ENOMEM to signal
2113 * userspace it may retry with a bigger buffer,
2114 * so we need to use something else for this case
2115 */
2116 buf = vzalloc(len);
2117 if (!buf)
2118 return -ENOBUFS;
2119
2120 max_irqs = len / sizeof(struct kvm_s390_irq);
2121
2122 if (gi->origin && gisa_get_ipm(gi->origin)) {
2123 for (i = 0; i <= MAX_ISC; i++) {
2124 if (n == max_irqs) {
2125 /* signal userspace to try again */
2126 ret = -ENOMEM;
2127 goto out_nolock;
2128 }
2129 if (gisa_tac_ipm_gisc(gi->origin, i)) {
2130 irq = (struct kvm_s390_irq *) &buf[n];
2131 irq->type = KVM_S390_INT_IO(1, 0, 0, 0);
2132 irq->u.io.io_int_word = isc_to_int_word(i);
2133 n++;
2134 }
2135 }
2136 }
2137 fi = &kvm->arch.float_int;
2138 spin_lock(&fi->lock);
2139 for (i = 0; i < FIRQ_LIST_COUNT; i++) {
2140 list_for_each_entry(inti, &fi->lists[i], list) {
2141 if (n == max_irqs) {
2142 /* signal userspace to try again */
2143 ret = -ENOMEM;
2144 goto out;
2145 }
2146 inti_to_irq(inti, &buf[n]);
2147 n++;
2148 }
2149 }
2150 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs)) {
2151 if (n == max_irqs) {
2152 /* signal userspace to try again */
2153 ret = -ENOMEM;
2154 goto out;
2155 }
2156 irq = (struct kvm_s390_irq *) &buf[n];
2157 irq->type = KVM_S390_INT_SERVICE;
2158 irq->u.ext = fi->srv_signal;
2159 n++;
2160 }
2161 if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
2162 if (n == max_irqs) {
2163 /* signal userspace to try again */
2164 ret = -ENOMEM;
2165 goto out;
2166 }
2167 irq = (struct kvm_s390_irq *) &buf[n];
2168 irq->type = KVM_S390_MCHK;
2169 irq->u.mchk = fi->mchk;
2170 n++;
2171}
2172
2173out:
2174 spin_unlock(&fi->lock);
2175out_nolock:
2176 if (!ret && n > 0) {
2177 if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
2178 ret = -EFAULT;
2179 }
2180 vfree(buf);
2181
2182 return ret < 0 ? ret : n;
2183}
2184
2185static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
2186{
2187 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2188 struct kvm_s390_ais_all ais;
2189
2190 if (attr->attr < sizeof(ais))
2191 return -EINVAL;
2192
2193 if (!test_kvm_facility(kvm, 72))
2194 return -ENOTSUPP;
2195
2196 mutex_lock(&fi->ais_lock);
2197 ais.simm = fi->simm;
2198 ais.nimm = fi->nimm;
2199 mutex_unlock(&fi->ais_lock);
2200
2201 if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
2202 return -EFAULT;
2203
2204 return 0;
2205}
2206
2207static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2208{
2209 int r;
2210
2211 switch (attr->group) {
2212 case KVM_DEV_FLIC_GET_ALL_IRQS:
2213 r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
2214 attr->attr);
2215 break;
2216 case KVM_DEV_FLIC_AISM_ALL:
2217 r = flic_ais_mode_get_all(dev->kvm, attr);
2218 break;
2219 default:
2220 r = -EINVAL;
2221 }
2222
2223 return r;
2224}
2225
2226static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
2227 u64 addr)
2228{
2229 struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
2230 void *target = NULL;
2231 void __user *source;
2232 u64 size;
2233
2234 if (get_user(inti->type, (u64 __user *)addr))
2235 return -EFAULT;
2236
2237 switch (inti->type) {
2238 case KVM_S390_INT_PFAULT_INIT:
2239 case KVM_S390_INT_PFAULT_DONE:
2240 case KVM_S390_INT_VIRTIO:
2241 case KVM_S390_INT_SERVICE:
2242 target = (void *) &inti->ext;
2243 source = &uptr->u.ext;
2244 size = sizeof(inti->ext);
2245 break;
2246 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2247 target = (void *) &inti->io;
2248 source = &uptr->u.io;
2249 size = sizeof(inti->io);
2250 break;
2251 case KVM_S390_MCHK:
2252 target = (void *) &inti->mchk;
2253 source = &uptr->u.mchk;
2254 size = sizeof(inti->mchk);
2255 break;
2256 default:
2257 return -EINVAL;
2258 }
2259
2260 if (copy_from_user(target, source, size))
2261 return -EFAULT;
2262
2263 return 0;
2264}
2265
2266static int enqueue_floating_irq(struct kvm_device *dev,
2267 struct kvm_device_attr *attr)
2268{
2269 struct kvm_s390_interrupt_info *inti = NULL;
2270 int r = 0;
2271 int len = attr->attr;
2272
2273 if (len % sizeof(struct kvm_s390_irq) != 0)
2274 return -EINVAL;
2275 else if (len > KVM_S390_FLIC_MAX_BUFFER)
2276 return -EINVAL;
2277
2278 while (len >= sizeof(struct kvm_s390_irq)) {
2279 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
2280 if (!inti)
2281 return -ENOMEM;
2282
2283 r = copy_irq_from_user(inti, attr->addr);
2284 if (r) {
2285 kfree(inti);
2286 return r;
2287 }
2288 r = __inject_vm(dev->kvm, inti);
2289 if (r) {
2290 kfree(inti);
2291 return r;
2292 }
2293 len -= sizeof(struct kvm_s390_irq);
2294 attr->addr += sizeof(struct kvm_s390_irq);
2295 }
2296
2297 return r;
2298}
2299
2300static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2301{
2302 if (id >= MAX_S390_IO_ADAPTERS)
2303 return NULL;
2304 id = array_index_nospec(id, MAX_S390_IO_ADAPTERS);
2305 return kvm->arch.adapters[id];
2306}
2307
2308static int register_io_adapter(struct kvm_device *dev,
2309 struct kvm_device_attr *attr)
2310{
2311 struct s390_io_adapter *adapter;
2312 struct kvm_s390_io_adapter adapter_info;
2313
2314 if (copy_from_user(&adapter_info,
2315 (void __user *)attr->addr, sizeof(adapter_info)))
2316 return -EFAULT;
2317
2318 if (adapter_info.id >= MAX_S390_IO_ADAPTERS)
2319 return -EINVAL;
2320
2321 adapter_info.id = array_index_nospec(adapter_info.id,
2322 MAX_S390_IO_ADAPTERS);
2323
2324 if (dev->kvm->arch.adapters[adapter_info.id] != NULL)
2325 return -EINVAL;
2326
2327 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2328 if (!adapter)
2329 return -ENOMEM;
2330
2331 INIT_LIST_HEAD(&adapter->maps);
2332 init_rwsem(&adapter->maps_lock);
2333 atomic_set(&adapter->nr_maps, 0);
2334 adapter->id = adapter_info.id;
2335 adapter->isc = adapter_info.isc;
2336 adapter->maskable = adapter_info.maskable;
2337 adapter->masked = false;
2338 adapter->swap = adapter_info.swap;
2339 adapter->suppressible = (adapter_info.flags) &
2340 KVM_S390_ADAPTER_SUPPRESSIBLE;
2341 dev->kvm->arch.adapters[adapter->id] = adapter;
2342
2343 return 0;
2344}
2345
2346int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2347{
2348 int ret;
2349 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2350
2351 if (!adapter || !adapter->maskable)
2352 return -EINVAL;
2353 ret = adapter->masked;
2354 adapter->masked = masked;
2355 return ret;
2356}
2357
2358static int kvm_s390_adapter_map(struct kvm *kvm, unsigned int id, __u64 addr)
2359{
2360 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2361 struct s390_map_info *map;
2362 int ret;
2363
2364 if (!adapter || !addr)
2365 return -EINVAL;
2366
2367 map = kzalloc(sizeof(*map), GFP_KERNEL);
2368 if (!map) {
2369 ret = -ENOMEM;
2370 goto out;
2371 }
2372 INIT_LIST_HEAD(&map->list);
2373 map->guest_addr = addr;
2374 map->addr = gmap_translate(kvm->arch.gmap, addr);
2375 if (map->addr == -EFAULT) {
2376 ret = -EFAULT;
2377 goto out;
2378 }
2379 ret = get_user_pages_fast(map->addr, 1, FOLL_WRITE, &map->page);
2380 if (ret < 0)
2381 goto out;
2382 BUG_ON(ret != 1);
2383 down_write(&adapter->maps_lock);
2384 if (atomic_inc_return(&adapter->nr_maps) < MAX_S390_ADAPTER_MAPS) {
2385 list_add_tail(&map->list, &adapter->maps);
2386 ret = 0;
2387 } else {
2388 put_page(map->page);
2389 ret = -EINVAL;
2390 }
2391 up_write(&adapter->maps_lock);
2392out:
2393 if (ret)
2394 kfree(map);
2395 return ret;
2396}
2397
2398static int kvm_s390_adapter_unmap(struct kvm *kvm, unsigned int id, __u64 addr)
2399{
2400 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2401 struct s390_map_info *map, *tmp;
2402 int found = 0;
2403
2404 if (!adapter || !addr)
2405 return -EINVAL;
2406
2407 down_write(&adapter->maps_lock);
2408 list_for_each_entry_safe(map, tmp, &adapter->maps, list) {
2409 if (map->guest_addr == addr) {
2410 found = 1;
2411 atomic_dec(&adapter->nr_maps);
2412 list_del(&map->list);
2413 put_page(map->page);
2414 kfree(map);
2415 break;
2416 }
2417 }
2418 up_write(&adapter->maps_lock);
2419
2420 return found ? 0 : -EINVAL;
2421}
2422
2423void kvm_s390_destroy_adapters(struct kvm *kvm)
2424{
2425 int i;
2426 struct s390_map_info *map, *tmp;
2427
2428 for (i = 0; i < MAX_S390_IO_ADAPTERS; i++) {
2429 if (!kvm->arch.adapters[i])
2430 continue;
2431 list_for_each_entry_safe(map, tmp,
2432 &kvm->arch.adapters[i]->maps, list) {
2433 list_del(&map->list);
2434 put_page(map->page);
2435 kfree(map);
2436 }
2437 kfree(kvm->arch.adapters[i]);
2438 }
2439}
2440
2441static int modify_io_adapter(struct kvm_device *dev,
2442 struct kvm_device_attr *attr)
2443{
2444 struct kvm_s390_io_adapter_req req;
2445 struct s390_io_adapter *adapter;
2446 int ret;
2447
2448 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2449 return -EFAULT;
2450
2451 adapter = get_io_adapter(dev->kvm, req.id);
2452 if (!adapter)
2453 return -EINVAL;
2454 switch (req.type) {
2455 case KVM_S390_IO_ADAPTER_MASK:
2456 ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2457 if (ret > 0)
2458 ret = 0;
2459 break;
2460 case KVM_S390_IO_ADAPTER_MAP:
2461 ret = kvm_s390_adapter_map(dev->kvm, req.id, req.addr);
2462 break;
2463 case KVM_S390_IO_ADAPTER_UNMAP:
2464 ret = kvm_s390_adapter_unmap(dev->kvm, req.id, req.addr);
2465 break;
2466 default:
2467 ret = -EINVAL;
2468 }
2469
2470 return ret;
2471}
2472
2473static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2474
2475{
2476 const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2477 u32 schid;
2478
2479 if (attr->flags)
2480 return -EINVAL;
2481 if (attr->attr != sizeof(schid))
2482 return -EINVAL;
2483 if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2484 return -EFAULT;
2485 if (!schid)
2486 return -EINVAL;
2487 kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2488 /*
2489 * If userspace is conforming to the architecture, we can have at most
2490 * one pending I/O interrupt per subchannel, so this is effectively a
2491 * clear all.
2492 */
2493 return 0;
2494}
2495
2496static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2497{
2498 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2499 struct kvm_s390_ais_req req;
2500 int ret = 0;
2501
2502 if (!test_kvm_facility(kvm, 72))
2503 return -ENOTSUPP;
2504
2505 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2506 return -EFAULT;
2507
2508 if (req.isc > MAX_ISC)
2509 return -EINVAL;
2510
2511 trace_kvm_s390_modify_ais_mode(req.isc,
2512 (fi->simm & AIS_MODE_MASK(req.isc)) ?
2513 (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2514 2 : KVM_S390_AIS_MODE_SINGLE :
2515 KVM_S390_AIS_MODE_ALL, req.mode);
2516
2517 mutex_lock(&fi->ais_lock);
2518 switch (req.mode) {
2519 case KVM_S390_AIS_MODE_ALL:
2520 fi->simm &= ~AIS_MODE_MASK(req.isc);
2521 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2522 break;
2523 case KVM_S390_AIS_MODE_SINGLE:
2524 fi->simm |= AIS_MODE_MASK(req.isc);
2525 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2526 break;
2527 default:
2528 ret = -EINVAL;
2529 }
2530 mutex_unlock(&fi->ais_lock);
2531
2532 return ret;
2533}
2534
2535static int kvm_s390_inject_airq(struct kvm *kvm,
2536 struct s390_io_adapter *adapter)
2537{
2538 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2539 struct kvm_s390_interrupt s390int = {
2540 .type = KVM_S390_INT_IO(1, 0, 0, 0),
2541 .parm = 0,
2542 .parm64 = isc_to_int_word(adapter->isc),
2543 };
2544 int ret = 0;
2545
2546 if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2547 return kvm_s390_inject_vm(kvm, &s390int);
2548
2549 mutex_lock(&fi->ais_lock);
2550 if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2551 trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2552 goto out;
2553 }
2554
2555 ret = kvm_s390_inject_vm(kvm, &s390int);
2556 if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2557 fi->nimm |= AIS_MODE_MASK(adapter->isc);
2558 trace_kvm_s390_modify_ais_mode(adapter->isc,
2559 KVM_S390_AIS_MODE_SINGLE, 2);
2560 }
2561out:
2562 mutex_unlock(&fi->ais_lock);
2563 return ret;
2564}
2565
2566static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2567{
2568 unsigned int id = attr->attr;
2569 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2570
2571 if (!adapter)
2572 return -EINVAL;
2573
2574 return kvm_s390_inject_airq(kvm, adapter);
2575}
2576
2577static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2578{
2579 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2580 struct kvm_s390_ais_all ais;
2581
2582 if (!test_kvm_facility(kvm, 72))
2583 return -ENOTSUPP;
2584
2585 if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2586 return -EFAULT;
2587
2588 mutex_lock(&fi->ais_lock);
2589 fi->simm = ais.simm;
2590 fi->nimm = ais.nimm;
2591 mutex_unlock(&fi->ais_lock);
2592
2593 return 0;
2594}
2595
2596static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2597{
2598 int r = 0;
2599 unsigned int i;
2600 struct kvm_vcpu *vcpu;
2601
2602 switch (attr->group) {
2603 case KVM_DEV_FLIC_ENQUEUE:
2604 r = enqueue_floating_irq(dev, attr);
2605 break;
2606 case KVM_DEV_FLIC_CLEAR_IRQS:
2607 kvm_s390_clear_float_irqs(dev->kvm);
2608 break;
2609 case KVM_DEV_FLIC_APF_ENABLE:
2610 dev->kvm->arch.gmap->pfault_enabled = 1;
2611 break;
2612 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2613 dev->kvm->arch.gmap->pfault_enabled = 0;
2614 /*
2615 * Make sure no async faults are in transition when
2616 * clearing the queues. So we don't need to worry
2617 * about late coming workers.
2618 */
2619 synchronize_srcu(&dev->kvm->srcu);
2620 kvm_for_each_vcpu(i, vcpu, dev->kvm)
2621 kvm_clear_async_pf_completion_queue(vcpu);
2622 break;
2623 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2624 r = register_io_adapter(dev, attr);
2625 break;
2626 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2627 r = modify_io_adapter(dev, attr);
2628 break;
2629 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2630 r = clear_io_irq(dev->kvm, attr);
2631 break;
2632 case KVM_DEV_FLIC_AISM:
2633 r = modify_ais_mode(dev->kvm, attr);
2634 break;
2635 case KVM_DEV_FLIC_AIRQ_INJECT:
2636 r = flic_inject_airq(dev->kvm, attr);
2637 break;
2638 case KVM_DEV_FLIC_AISM_ALL:
2639 r = flic_ais_mode_set_all(dev->kvm, attr);
2640 break;
2641 default:
2642 r = -EINVAL;
2643 }
2644
2645 return r;
2646}
2647
2648static int flic_has_attr(struct kvm_device *dev,
2649 struct kvm_device_attr *attr)
2650{
2651 switch (attr->group) {
2652 case KVM_DEV_FLIC_GET_ALL_IRQS:
2653 case KVM_DEV_FLIC_ENQUEUE:
2654 case KVM_DEV_FLIC_CLEAR_IRQS:
2655 case KVM_DEV_FLIC_APF_ENABLE:
2656 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2657 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2658 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2659 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2660 case KVM_DEV_FLIC_AISM:
2661 case KVM_DEV_FLIC_AIRQ_INJECT:
2662 case KVM_DEV_FLIC_AISM_ALL:
2663 return 0;
2664 }
2665 return -ENXIO;
2666}
2667
2668static int flic_create(struct kvm_device *dev, u32 type)
2669{
2670 if (!dev)
2671 return -EINVAL;
2672 if (dev->kvm->arch.flic)
2673 return -EINVAL;
2674 dev->kvm->arch.flic = dev;
2675 return 0;
2676}
2677
2678static void flic_destroy(struct kvm_device *dev)
2679{
2680 dev->kvm->arch.flic = NULL;
2681 kfree(dev);
2682}
2683
2684/* s390 floating irq controller (flic) */
2685struct kvm_device_ops kvm_flic_ops = {
2686 .name = "kvm-flic",
2687 .get_attr = flic_get_attr,
2688 .set_attr = flic_set_attr,
2689 .has_attr = flic_has_attr,
2690 .create = flic_create,
2691 .destroy = flic_destroy,
2692};
2693
2694static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2695{
2696 unsigned long bit;
2697
2698 bit = bit_nr + (addr % PAGE_SIZE) * 8;
2699
2700 return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2701}
2702
2703static struct s390_map_info *get_map_info(struct s390_io_adapter *adapter,
2704 u64 addr)
2705{
2706 struct s390_map_info *map;
2707
2708 if (!adapter)
2709 return NULL;
2710
2711 list_for_each_entry(map, &adapter->maps, list) {
2712 if (map->guest_addr == addr)
2713 return map;
2714 }
2715 return NULL;
2716}
2717
2718static int adapter_indicators_set(struct kvm *kvm,
2719 struct s390_io_adapter *adapter,
2720 struct kvm_s390_adapter_int *adapter_int)
2721{
2722 unsigned long bit;
2723 int summary_set, idx;
2724 struct s390_map_info *info;
2725 void *map;
2726
2727 info = get_map_info(adapter, adapter_int->ind_addr);
2728 if (!info)
2729 return -1;
2730 map = page_address(info->page);
2731 bit = get_ind_bit(info->addr, adapter_int->ind_offset, adapter->swap);
2732 set_bit(bit, map);
2733 idx = srcu_read_lock(&kvm->srcu);
2734 mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2735 set_page_dirty_lock(info->page);
2736 info = get_map_info(adapter, adapter_int->summary_addr);
2737 if (!info) {
2738 srcu_read_unlock(&kvm->srcu, idx);
2739 return -1;
2740 }
2741 map = page_address(info->page);
2742 bit = get_ind_bit(info->addr, adapter_int->summary_offset,
2743 adapter->swap);
2744 summary_set = test_and_set_bit(bit, map);
2745 mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2746 set_page_dirty_lock(info->page);
2747 srcu_read_unlock(&kvm->srcu, idx);
2748 return summary_set ? 0 : 1;
2749}
2750
2751/*
2752 * < 0 - not injected due to error
2753 * = 0 - coalesced, summary indicator already active
2754 * > 0 - injected interrupt
2755 */
2756static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2757 struct kvm *kvm, int irq_source_id, int level,
2758 bool line_status)
2759{
2760 int ret;
2761 struct s390_io_adapter *adapter;
2762
2763 /* We're only interested in the 0->1 transition. */
2764 if (!level)
2765 return 0;
2766 adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2767 if (!adapter)
2768 return -1;
2769 down_read(&adapter->maps_lock);
2770 ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2771 up_read(&adapter->maps_lock);
2772 if ((ret > 0) && !adapter->masked) {
2773 ret = kvm_s390_inject_airq(kvm, adapter);
2774 if (ret == 0)
2775 ret = 1;
2776 }
2777 return ret;
2778}
2779
2780/*
2781 * Inject the machine check to the guest.
2782 */
2783void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2784 struct mcck_volatile_info *mcck_info)
2785{
2786 struct kvm_s390_interrupt_info inti;
2787 struct kvm_s390_irq irq;
2788 struct kvm_s390_mchk_info *mchk;
2789 union mci mci;
2790 __u64 cr14 = 0; /* upper bits are not used */
2791 int rc;
2792
2793 mci.val = mcck_info->mcic;
2794 if (mci.sr)
2795 cr14 |= CR14_RECOVERY_SUBMASK;
2796 if (mci.dg)
2797 cr14 |= CR14_DEGRADATION_SUBMASK;
2798 if (mci.w)
2799 cr14 |= CR14_WARNING_SUBMASK;
2800
2801 mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2802 mchk->cr14 = cr14;
2803 mchk->mcic = mcck_info->mcic;
2804 mchk->ext_damage_code = mcck_info->ext_damage_code;
2805 mchk->failing_storage_address = mcck_info->failing_storage_address;
2806 if (mci.ck) {
2807 /* Inject the floating machine check */
2808 inti.type = KVM_S390_MCHK;
2809 rc = __inject_vm(vcpu->kvm, &inti);
2810 } else {
2811 /* Inject the machine check to specified vcpu */
2812 irq.type = KVM_S390_MCHK;
2813 rc = kvm_s390_inject_vcpu(vcpu, &irq);
2814 }
2815 WARN_ON_ONCE(rc);
2816}
2817
2818int kvm_set_routing_entry(struct kvm *kvm,
2819 struct kvm_kernel_irq_routing_entry *e,
2820 const struct kvm_irq_routing_entry *ue)
2821{
2822 int ret;
2823
2824 switch (ue->type) {
2825 case KVM_IRQ_ROUTING_S390_ADAPTER:
2826 e->set = set_adapter_int;
2827 e->adapter.summary_addr = ue->u.adapter.summary_addr;
2828 e->adapter.ind_addr = ue->u.adapter.ind_addr;
2829 e->adapter.summary_offset = ue->u.adapter.summary_offset;
2830 e->adapter.ind_offset = ue->u.adapter.ind_offset;
2831 e->adapter.adapter_id = ue->u.adapter.adapter_id;
2832 ret = 0;
2833 break;
2834 default:
2835 ret = -EINVAL;
2836 }
2837
2838 return ret;
2839}
2840
2841int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2842 int irq_source_id, int level, bool line_status)
2843{
2844 return -EINVAL;
2845}
2846
2847int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2848{
2849 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2850 struct kvm_s390_irq *buf;
2851 int r = 0;
2852 int n;
2853
2854 buf = vmalloc(len);
2855 if (!buf)
2856 return -ENOMEM;
2857
2858 if (copy_from_user((void *) buf, irqstate, len)) {
2859 r = -EFAULT;
2860 goto out_free;
2861 }
2862
2863 /*
2864 * Don't allow setting the interrupt state
2865 * when there are already interrupts pending
2866 */
2867 spin_lock(&li->lock);
2868 if (li->pending_irqs) {
2869 r = -EBUSY;
2870 goto out_unlock;
2871 }
2872
2873 for (n = 0; n < len / sizeof(*buf); n++) {
2874 r = do_inject_vcpu(vcpu, &buf[n]);
2875 if (r)
2876 break;
2877 }
2878
2879out_unlock:
2880 spin_unlock(&li->lock);
2881out_free:
2882 vfree(buf);
2883
2884 return r;
2885}
2886
2887static void store_local_irq(struct kvm_s390_local_interrupt *li,
2888 struct kvm_s390_irq *irq,
2889 unsigned long irq_type)
2890{
2891 switch (irq_type) {
2892 case IRQ_PEND_MCHK_EX:
2893 case IRQ_PEND_MCHK_REP:
2894 irq->type = KVM_S390_MCHK;
2895 irq->u.mchk = li->irq.mchk;
2896 break;
2897 case IRQ_PEND_PROG:
2898 irq->type = KVM_S390_PROGRAM_INT;
2899 irq->u.pgm = li->irq.pgm;
2900 break;
2901 case IRQ_PEND_PFAULT_INIT:
2902 irq->type = KVM_S390_INT_PFAULT_INIT;
2903 irq->u.ext = li->irq.ext;
2904 break;
2905 case IRQ_PEND_EXT_EXTERNAL:
2906 irq->type = KVM_S390_INT_EXTERNAL_CALL;
2907 irq->u.extcall = li->irq.extcall;
2908 break;
2909 case IRQ_PEND_EXT_CLOCK_COMP:
2910 irq->type = KVM_S390_INT_CLOCK_COMP;
2911 break;
2912 case IRQ_PEND_EXT_CPU_TIMER:
2913 irq->type = KVM_S390_INT_CPU_TIMER;
2914 break;
2915 case IRQ_PEND_SIGP_STOP:
2916 irq->type = KVM_S390_SIGP_STOP;
2917 irq->u.stop = li->irq.stop;
2918 break;
2919 case IRQ_PEND_RESTART:
2920 irq->type = KVM_S390_RESTART;
2921 break;
2922 case IRQ_PEND_SET_PREFIX:
2923 irq->type = KVM_S390_SIGP_SET_PREFIX;
2924 irq->u.prefix = li->irq.prefix;
2925 break;
2926 }
2927}
2928
2929int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
2930{
2931 int scn;
2932 DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
2933 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2934 unsigned long pending_irqs;
2935 struct kvm_s390_irq irq;
2936 unsigned long irq_type;
2937 int cpuaddr;
2938 int n = 0;
2939
2940 spin_lock(&li->lock);
2941 pending_irqs = li->pending_irqs;
2942 memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
2943 sizeof(sigp_emerg_pending));
2944 spin_unlock(&li->lock);
2945
2946 for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
2947 memset(&irq, 0, sizeof(irq));
2948 if (irq_type == IRQ_PEND_EXT_EMERGENCY)
2949 continue;
2950 if (n + sizeof(irq) > len)
2951 return -ENOBUFS;
2952 store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
2953 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2954 return -EFAULT;
2955 n += sizeof(irq);
2956 }
2957
2958 if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
2959 for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
2960 memset(&irq, 0, sizeof(irq));
2961 if (n + sizeof(irq) > len)
2962 return -ENOBUFS;
2963 irq.type = KVM_S390_INT_EMERGENCY;
2964 irq.u.emerg.code = cpuaddr;
2965 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2966 return -EFAULT;
2967 n += sizeof(irq);
2968 }
2969 }
2970
2971 if (sca_ext_call_pending(vcpu, &scn)) {
2972 if (n + sizeof(irq) > len)
2973 return -ENOBUFS;
2974 memset(&irq, 0, sizeof(irq));
2975 irq.type = KVM_S390_INT_EXTERNAL_CALL;
2976 irq.u.extcall.code = scn;
2977 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2978 return -EFAULT;
2979 n += sizeof(irq);
2980 }
2981
2982 return n;
2983}
2984
2985static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
2986{
2987 int vcpu_id, online_vcpus = atomic_read(&kvm->online_vcpus);
2988 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2989 struct kvm_vcpu *vcpu;
2990
2991 for_each_set_bit(vcpu_id, kvm->arch.idle_mask, online_vcpus) {
2992 vcpu = kvm_get_vcpu(kvm, vcpu_id);
2993 if (psw_ioint_disabled(vcpu))
2994 continue;
2995 deliverable_mask &= (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
2996 if (deliverable_mask) {
2997 /* lately kicked but not yet running */
2998 if (test_and_set_bit(vcpu_id, gi->kicked_mask))
2999 return;
3000 kvm_s390_vcpu_wakeup(vcpu);
3001 return;
3002 }
3003 }
3004}
3005
3006static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3007{
3008 struct kvm_s390_gisa_interrupt *gi =
3009 container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3010 struct kvm *kvm =
3011 container_of(gi->origin, struct sie_page2, gisa)->kvm;
3012 u8 pending_mask;
3013
3014 pending_mask = gisa_get_ipm_or_restore_iam(gi);
3015 if (pending_mask) {
3016 __airqs_kick_single_vcpu(kvm, pending_mask);
3017 hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3018 return HRTIMER_RESTART;
3019 };
3020
3021 return HRTIMER_NORESTART;
3022}
3023
3024#define NULL_GISA_ADDR 0x00000000UL
3025#define NONE_GISA_ADDR 0x00000001UL
3026#define GISA_ADDR_MASK 0xfffff000UL
3027
3028static void process_gib_alert_list(void)
3029{
3030 struct kvm_s390_gisa_interrupt *gi;
3031 struct kvm_s390_gisa *gisa;
3032 struct kvm *kvm;
3033 u32 final, origin = 0UL;
3034
3035 do {
3036 /*
3037 * If the NONE_GISA_ADDR is still stored in the alert list
3038 * origin, we will leave the outer loop. No further GISA has
3039 * been added to the alert list by millicode while processing
3040 * the current alert list.
3041 */
3042 final = (origin & NONE_GISA_ADDR);
3043 /*
3044 * Cut off the alert list and store the NONE_GISA_ADDR in the
3045 * alert list origin to avoid further GAL interruptions.
3046 * A new alert list can be build up by millicode in parallel
3047 * for guests not in the yet cut-off alert list. When in the
3048 * final loop, store the NULL_GISA_ADDR instead. This will re-
3049 * enable GAL interruptions on the host again.
3050 */
3051 origin = xchg(&gib->alert_list_origin,
3052 (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3053 /*
3054 * Loop through the just cut-off alert list and start the
3055 * gisa timers to kick idle vcpus to consume the pending
3056 * interruptions asap.
3057 */
3058 while (origin & GISA_ADDR_MASK) {
3059 gisa = (struct kvm_s390_gisa *)(u64)origin;
3060 origin = gisa->next_alert;
3061 gisa->next_alert = (u32)(u64)gisa;
3062 kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3063 gi = &kvm->arch.gisa_int;
3064 if (hrtimer_active(&gi->timer))
3065 hrtimer_cancel(&gi->timer);
3066 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3067 }
3068 } while (!final);
3069
3070}
3071
3072void kvm_s390_gisa_clear(struct kvm *kvm)
3073{
3074 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3075
3076 if (!gi->origin)
3077 return;
3078 gisa_clear_ipm(gi->origin);
3079 VM_EVENT(kvm, 3, "gisa 0x%pK cleared", gi->origin);
3080}
3081
3082void kvm_s390_gisa_init(struct kvm *kvm)
3083{
3084 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3085
3086 if (!css_general_characteristics.aiv)
3087 return;
3088 gi->origin = &kvm->arch.sie_page2->gisa;
3089 gi->alert.mask = 0;
3090 spin_lock_init(&gi->alert.ref_lock);
3091 gi->expires = 50 * 1000; /* 50 usec */
3092 hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3093 gi->timer.function = gisa_vcpu_kicker;
3094 memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3095 gi->origin->next_alert = (u32)(u64)gi->origin;
3096 VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin);
3097}
3098
3099void kvm_s390_gisa_destroy(struct kvm *kvm)
3100{
3101 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3102
3103 if (!gi->origin)
3104 return;
3105 if (gi->alert.mask)
3106 KVM_EVENT(3, "vm 0x%pK has unexpected iam 0x%02x",
3107 kvm, gi->alert.mask);
3108 while (gisa_in_alert_list(gi->origin))
3109 cpu_relax();
3110 hrtimer_cancel(&gi->timer);
3111 gi->origin = NULL;
3112}
3113
3114/**
3115 * kvm_s390_gisc_register - register a guest ISC
3116 *
3117 * @kvm: the kernel vm to work with
3118 * @gisc: the guest interruption sub class to register
3119 *
3120 * The function extends the vm specific alert mask to use.
3121 * The effective IAM mask in the GISA is updated as well
3122 * in case the GISA is not part of the GIB alert list.
3123 * It will be updated latest when the IAM gets restored
3124 * by gisa_get_ipm_or_restore_iam().
3125 *
3126 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3127 * has registered with the channel subsystem.
3128 * -ENODEV in case the vm uses no GISA
3129 * -ERANGE in case the guest ISC is invalid
3130 */
3131int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3132{
3133 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3134
3135 if (!gi->origin)
3136 return -ENODEV;
3137 if (gisc > MAX_ISC)
3138 return -ERANGE;
3139
3140 spin_lock(&gi->alert.ref_lock);
3141 gi->alert.ref_count[gisc]++;
3142 if (gi->alert.ref_count[gisc] == 1) {
3143 gi->alert.mask |= 0x80 >> gisc;
3144 gisa_set_iam(gi->origin, gi->alert.mask);
3145 }
3146 spin_unlock(&gi->alert.ref_lock);
3147
3148 return gib->nisc;
3149}
3150EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3151
3152/**
3153 * kvm_s390_gisc_unregister - unregister a guest ISC
3154 *
3155 * @kvm: the kernel vm to work with
3156 * @gisc: the guest interruption sub class to register
3157 *
3158 * The function reduces the vm specific alert mask to use.
3159 * The effective IAM mask in the GISA is updated as well
3160 * in case the GISA is not part of the GIB alert list.
3161 * It will be updated latest when the IAM gets restored
3162 * by gisa_get_ipm_or_restore_iam().
3163 *
3164 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3165 * has registered with the channel subsystem.
3166 * -ENODEV in case the vm uses no GISA
3167 * -ERANGE in case the guest ISC is invalid
3168 * -EINVAL in case the guest ISC is not registered
3169 */
3170int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3171{
3172 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3173 int rc = 0;
3174
3175 if (!gi->origin)
3176 return -ENODEV;
3177 if (gisc > MAX_ISC)
3178 return -ERANGE;
3179
3180 spin_lock(&gi->alert.ref_lock);
3181 if (gi->alert.ref_count[gisc] == 0) {
3182 rc = -EINVAL;
3183 goto out;
3184 }
3185 gi->alert.ref_count[gisc]--;
3186 if (gi->alert.ref_count[gisc] == 0) {
3187 gi->alert.mask &= ~(0x80 >> gisc);
3188 gisa_set_iam(gi->origin, gi->alert.mask);
3189 }
3190out:
3191 spin_unlock(&gi->alert.ref_lock);
3192
3193 return rc;
3194}
3195EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3196
3197static void gib_alert_irq_handler(struct airq_struct *airq, bool floating)
3198{
3199 inc_irq_stat(IRQIO_GAL);
3200 process_gib_alert_list();
3201}
3202
3203static struct airq_struct gib_alert_irq = {
3204 .handler = gib_alert_irq_handler,
3205 .lsi_ptr = &gib_alert_irq.lsi_mask,
3206};
3207
3208void kvm_s390_gib_destroy(void)
3209{
3210 if (!gib)
3211 return;
3212 chsc_sgib(0);
3213 unregister_adapter_interrupt(&gib_alert_irq);
3214 free_page((unsigned long)gib);
3215 gib = NULL;
3216}
3217
3218int kvm_s390_gib_init(u8 nisc)
3219{
3220 int rc = 0;
3221
3222 if (!css_general_characteristics.aiv) {
3223 KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3224 goto out;
3225 }
3226
3227 gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL | GFP_DMA);
3228 if (!gib) {
3229 rc = -ENOMEM;
3230 goto out;
3231 }
3232
3233 gib_alert_irq.isc = nisc;
3234 if (register_adapter_interrupt(&gib_alert_irq)) {
3235 pr_err("Registering the GIB alert interruption handler failed\n");
3236 rc = -EIO;
3237 goto out_free_gib;
3238 }
3239
3240 gib->nisc = nisc;
3241 if (chsc_sgib((u32)(u64)gib)) {
3242 pr_err("Associating the GIB with the AIV facility failed\n");
3243 free_page((unsigned long)gib);
3244 gib = NULL;
3245 rc = -EIO;
3246 goto out_unreg_gal;
3247 }
3248
3249 KVM_EVENT(3, "gib 0x%pK (nisc=%d) initialized", gib, gib->nisc);
3250 goto out;
3251
3252out_unreg_gal:
3253 unregister_adapter_interrupt(&gib_alert_irq);
3254out_free_gib:
3255 free_page((unsigned long)gib);
3256 gib = NULL;
3257out:
3258 return rc;
3259}